Formosan Sambar Deer Space Use and Multiscale Habitat Selection Using Habitat Suitability Modelling and GPS Telemetry

國立臺灣師範大學生命科學系博士論文

以棲地適合度模式與 GPS 遙測技術探討臺灣水鹿之空間使用及不同尺度下之棲地選擇方式 Formosan sambar deer space use and multiscale habitat selection using habitat suitability modelling and GPS telemetry

研究生：顏士清 Shih-Ching Yen

指導教授：王穎博士 Ying Wang

中華民國 102 年 7 月

Table of Contents Pages

摘要 1

Abstract 3

Introduction 5

Methods 10

Results 23

Discussion 27

Management implications 38

Conclusion 40

References 41

List of Tables

Pages

Table 1 The elevational distribution of Formosan sambar deer (Rusa 55

unicolor swinhoii) presence-absence records in Taiwan.

Table 2 List of environmental variables used to predict the distribution of 56

Formosan sambar deer (Rusa unicolor swinhoii) in Taiwan.

Table 3 Ranks of gain contributions of environmental variables in the 4 57

habitat suitability models: logistic regression, discriminant analysis,

Ecological-Niche Factor Analysis (ENFA), and Maximum Entropy

(Maxent). The other habitat suitability model, Genetic Algorithm for

Rule-set Production, could not be used to compare the gain

contributions of each variable. ENFA could not be used to compute

nominal variables; therefore, the variable “vegetation type” was

excluded from ENFA computation.

Table 4 Data describing locations collected from Formosan sambar deer 58

with GPS collars in Taroko National Park, Taiwan, from December

2009 to July 2013.

Table 5 Properties of the habitat types used to analyze habitat selection by 59

Formosan sambar deer in Taroko National Park, Taiwan, from

December 2009 to July 2013.

Table 6 Second-order habitat selection (home-range scale) by individual 60

deer (n=12) by using landscape attributes in a Formosan sambar deer

GPS telemetry study in Taroko National Park, Taiwan, from

December 2009 to July 2013. “Number of deer ” reflects the number

of individual deer with mean values greater than (+), less than (-), or

equal to (0) the mean values of study area, based on significant t-test

(for slope, elevation, and solar duration) (P < 0.05) and Bonferroni

confidence interval (for aspect) between home range and study area.

Table 7 Second-order habitat selection (home-range scale) of habitat type 61

determined by Euclidean distance analysis using location data by 12

Formosan sambar deer in Taroko National Park, Taiwan, from

December 2009 to July 2013.

Table 8 Third-order habitat selection (within-home-range sacle) by 62

individual deer (n=12) by using landscape attributes in a Formosan

sambar deer GPS telemetry study in Taroko National Park, Taiwan,

from December 2009 to July 2013. “Number of deer ” reflects the

number of individual deer with mean use values greater than (+), less

than (-), or equal to (0) the mean value of an associated set of random

locations, based on significant t-test (for slope, elevation, and solar

duration) (P < 0.05) and Bonferroni confidence interval (for aspect)

between use and random locations.

Table 9 Third-order habitat selection (within-home-range scale) of habitat 63

type determined by Euclidean distance analysis using location data by

12 Formosan sambar deer in Taroko National Park, Taiwan, from

December 2009 to July 2013.

Table 10 Comparison of the distances (m) from forest pathes between 64

daytime and night locations by 12 collared Formosan sambar deer in

Taroko National Park.

Table 11 Summary statistics for 100% minimum convex polygon (MCP) 65

and 95% fixed-kernel home ranges (ha) of Formosan sambar deer

estimated by GPS telemetry in Taroko National Park, Taiwan, from

December 2009 to July 2013.

Table 12 Summary statistics for daily displacement of Formosan sambar 66

deer estimated by GPS telemetry in Taroko National Park, Taiwan,

from December 2009 to July 2013.

List of Figures

Pages

Figure 1 (a) Elevation map of Taiwan, showing the locations of the 67

mountain ranges; (b) Distribution of protected areas across Taiwan.

Figure 2 Predicted habitat of Formosan sambar deer (Rusa unicolor 68

swinhoii) in Taiwan by using logistic regression, discriminant

analysis, Ecological-Niche Factor Analysis (ENFA), Genetic

Algorithm for Rule-set Production (GARP), and Maximum Entropy

(Maxent).

Figure 3 (a) Predicted habitat of Formosan sambar deer (Rusa unicolor 69

swinhoii) in Taiwan by using ensemble forecasting. There are 3

highways crossing the main habitats that are suitable for sambar

deer: Central Cross-Island Highway, Southern Cross-Island

Highway, and Highway No. 7A; (b) Recorded locations of Formosan

sambar deer (Rusa unicolor swinhoii) in Taiwan. Data were collected

from field surveys (2008–2011) and assimilated previous studies

(2002–2007).

Figure 4 Map of the study area showing the habitat types in the habitat 70

selection analysis. The study area was 100% minimum convex

polygon generated by all locations from 12 Formosan sambar deer

tracked between December 2009 and July 2013.

Figure 5 The mean and range of slopes used by collared Formosan sambar 71

deer in Taroko National Park, Taiwan, from December 2009 to July

2013.

Figure 6 Mean elevations of the collared Formosan sambar deer locations 72

during cold/dry season (November to April) and hot/wet season

(May to October) in Taroko National Park, Taiwan, from December

2009 to July 2013, shown as mean ± SD.

Figure 7 The proportions of aspects used by 12 Formosan sambar deer 73

during cold/dry season (November to April) and hot/wet season

(May to October) in Taroko National Park, Taiwan, from December

2009 to July 2013, shown as mean ± SD. The aspects were

reclassified into 3 moisture gradients: mesic (338-67°), subxeric

(68-157° and 248-337°), and xeric (158-247°).

Figure 8 The100% minimum convex polygon home ranges for 12 collared 74

Formosan sambar deer in Taroko National Park. The individuals

tracked at the same period were shown together at the same partition.

Figure 9 Home range overlap proportions among collared Formosan sambar 75

deer individuals. Overlap was computed as individual interactions

(see ’Methods’).

謝誌

五年的博士班生涯似乎很短暫，在我越來越發覺自己懂得太少時，學生身份

轉瞬間就結束了，但它又似乎很漫長，成天躲在山裡與動物為伍，令我的人生步

調與身邊的朋友都不太一致，我是很幸運的，這五年來積累的回憶與經歷絕非常

人所能體驗、體會。

父親在我小時候帶我踏進山林，總是鼓勵我”做點跟別人不一樣的事”，希望

我能讓您感到驕傲，也希望在天上的您有看到這一切。非常感謝王穎老師，指導

我獲得獨立思考與獨立研究的能力，在我博士班期間給了我最大程度的自由與支

持，我總是可以自由地安排時間、行程，研究背後又有穩定的經費在支持。謝謝

我的家人們，母親、哥哥總是默默支持我，從不質疑我做想做的事，你們是我最

倚賴的支柱。總是陪在我身邊、最了解我的人是筱筠，謝謝妳的等待與鼓勵，我

很幸運可以與妳結為夫妻。

進出奇萊山區二十餘趟，獲得的幫助實在可以洋洋灑灑寫出一篇直逼論文本

文的謝誌了，因此請恕我以較簡略的方式述說，我的感激難以用文字表達。五年

來的研究經費全靠太魯閣國家公園保育課的研究計劃支持，感謝陳俊山課長熱心

推動水鹿相關研究。捕捉水鹿的過程，衝在最前線拚搏的是一群勇士，包括經驗

豐富的資深布農獵人 Umas、Lingav，教導我們陷阱的架設與捕捉保定方法，一

群志同道合的好友：賴冠榮、陳匡洵、廖昱銓、林子揚、王立豪、張國威、許詩

涵、蔡南益、顏鴻榆、汪仁傑、楊宇帆，以及來自不同部落的布農勇士：高瑩山、

高新興、全志翔、何隼、Abis、鐵馬。捕捉工作中，不可或缺的還有獸醫的加入：

曾美萍、朱何宗、余品奐、毛祈鈞、吳盈慧、官苑芃、陳俊有、吳志純、陳家容、

顏鉅宇、陳儒頎、劉晉嘉、蔡伊婷、吳雯鈴、謝珮瑜，感謝你們暫時拋開工作或

課業跟我來到山上，讓水鹿與研究人員的安全雙雙獲得保障。捕捉的背後還需要

一群人的支援，協助器材準備、記錄、運送物資等，感謝王韋政、林欣怡、高嘉

孜、林祐竹、蔡佩芳、潘玉潔、蔡佳容、林欣儀、何紋靈、韓建國、何鑫、黃致

豪、李昀蒨、高詩豪、林玉佩、鄭淑如、郭正彥、楊琬菁、陳懿文、朱有田、白

欽源、陳佳利、陳添寶等人的熱心參與及幫忙。捕捉後其實才是辛苦的開始，還

有一群人陪著我東奔西跑追蹤水鹿：陳俞佑、程宗德、陸可凡、廖家宏、呂政翰、

莊又澄、黃敏琪、王郁傑、方翔，感謝你們協助這段辛苦的追蹤工作。

我受到的幫助不只在山上，山下還獲得好多其他支援，感謝歐恆佑大力協助

GIS 相關研究技術、林宗以提供水鹿分布資料，謝謝邱惠儀、陳怡君、呂翊維、

林致鋼等實驗室夥伴的後勤支援，感謝郭俊成、祈偉廉兩位獸醫師熱心指導麻醉

相關技術，還有屏科大野生動物收容中心、季昭華老師研究室、吳永惠老師研究

室提供獸醫藥品與器材的支援，在研究初期還受潘明雄及蔡木生指導陷阱架設技

術，及李培芬老師提供研究方法建議，並感謝墾丁國家公園管理處提供捕捉網幫

助研究的進行。

最後謝謝李玲玲老師、裴家騏老師、吳海音老師、謝寶森老師對這個研究的

細心指正，短短兩個小時的口試讓我學到很多，雖然目前還無法修改到盡善盡美，

在未來我仍會繼續努力讓研究更完整而正確。

摘要

欲擬訂有效的野生動物保育與經營管理策略，必須先了解動物的空間使用與

棲地選擇。臺灣水鹿(Rusa unicolor swinhoii)由於遭受到棲息地的破壞以及狩獵的

壓力，被列為保育類野生動物，但目前對其族群狀況及生活史的了解仍十分缺乏，

因此本研究首先以棲地適合度模式(habitat suitability modelling)探討其在臺灣的

分布情形，並找出影響其分布的關鍵因子，接著利用全球定位系統(global positioning system, GPS)追蹤技術，探討水鹿在活動範圍尺度(home-range scale)、

活動範圍內尺度(within-home-range scale)、時間尺度(temporal scale)的棲地選擇，

並了解其空間使用方式。棲地適合度模式顯示水鹿偏好棲息在海拔高於 1,500 m

及遠離公路的地區，臺灣具有 7,865 km2 適合水鹿生存的棲地，這些適合的棲地

主要分布在中央山脈與雪山山脈，但被三條山區省道切割為五個主要區塊，我們

建議監測部分靠近公路的水鹿棲地，未來可嘗試協助其建立亞族群間的交流。此

外，我們於 2009 年 7 月至 2013 年 7 月間藉由 GPS 項圈追蹤了 12 隻水鹿(6 雄 6

雌)，發現水鹿具有季節性移動行為，冷乾季(11 月到 4 月)棲息在海拔較低的地

區(平均 2,483 ± 406 m)，熱濕季(5 月到 10 月)移動到海拔較高的地區(平均 2,984 ±

222 m)。在活動範圍尺度，水鹿偏好使用較平坦的地形及較潮溼的坡向(338-67°)，

在冷乾季偏好使用闊葉林、針闊葉混淆林、開闊地、及鐵杉林，在熱濕季則偏好

冷杉林、箭竹草原、及鐵杉林，整體而言，水鹿能夠廣泛適應各種植被類型。在

活動範圍內尺度，水鹿在冷乾季偏好使用太陽照射時數較高的地點。水鹿的棲地

選擇也發生在時間尺度上，其在日間距森林的距離明顯短於夜間，顯示森林為水

鹿重要的掩蔽處所。在空間使用方面，以最小凸多邊形法(minimum convex polygon)評估水鹿活動範圍大小，發現雄鹿平均年活動範圍為 1,078 ± 501 ha，雌

鹿平均年活動範圍為 1,001 ± 346 ha，活動範圍間的重疊度可高達 80.2%，顯示水

鹿並未建立排他的領域。雄鹿平均日位移 268 ± 90 m，最大日位移 6,435 m，雌

鹿平均日位移 317 ± 135 m，最大日位移 4,422 m。活動範圍與日位移在性別間、

季節間都沒有顯著差異。整體而言，我們建立了水鹿棲地選擇與空間使用的完整

資訊，並推測未來水鹿族群擴張的主要限制是人為干擾的相關因子而非自然環境

因子。

關鍵字：地理資訊系統、棲地選擇、活動範圍、水鹿、野生動物經營管理

Abstract Studies on the animal space use and habitat selection are required for the conservation and management of large herbivores. In Taiwan, the Formosan sambar deer (Rusa unicolor swinhoii) is listed as a protected species under the wildlife conservation law because of human overexploitation. However, its population status and life history remains unclear. In this study, we used 2 approaches to investigate habitat selection and space use of sambar deer. In the study on geographical-range scale habitat selection, we used habitat suitability modelling to identify key habitat variables and to predict potential distribution of this species throughout Taiwan. In the studies on space use and habitat selections at home-range scale, within-home-range scale, and temporal scale, we tracked the deer by using global positioning system telemetry. The habitat suitability models indicated the presence of 7,865 km2 suitable habitats for the sambar deer in Taiwan. The deer preferentially used areas that were over 1,500 m in elevation and were distant from roads. The results predicted that deer habitats are mainly located in the Central Mountain Range and Xue Mountain Range of Taiwan. However, the predicted habitats were divided into 5 regions, which were separated by 3 major mountain highways. We recommend that deer hotspots close to the highways should be monitored for the future establishment of connections among different Formosan sambar deer sub-populations. Furthermore, we collected location data from 12 collared deer (6 males and 6 females) between December 2009 and July

2013. The collared deer used higher elevation areas in hot/wet season (mean: 2,984 ±

222 m; from May to October) than in cold/dry season (mean: 2,483 ± 406 m; from

November to April), which indicated a seasonal movement behavior. At the home-range scale, the deer preferred broadleaf forest, mixed forest, open habitat, and hemlock forest in the cold/dry season, and preferred fir forest, grassland, and hemlock forest in the hot/wet season. We suggested that sambar deer is a habitat generalist. In

addition, the deer preferred to use areas with flatter slopes and mesic aspects (338-67°) at home-range scale, and selected areas with higher solar duration at within-home range scale in the cold/dry season. Habitat selection also occurred at temporal scale.

The deer was usually closer to forested habitat in daytime than at night, highlighting its requirement of forest as shelter and cover. Moreover, the mean annual 100% minimum convex polygon home ranges were 1,078 ± 501 ha for males and 1,001 ±

346 ha for females. Overlap in home ranges of sambar deer could be as high as 80.2%, which suggested that the sambar deer did not establish exclusive territory. The mean daily displacements were 268 ± 90 m for males and 317 ± 135 m for females, with a maximum distance of 6,435 m for male and 4,422 m for female. No significant differences of home range size and daily displacement were detected between two sexes and between two seasons. In conclusion, we comprehensively studied the space use and habitat selection of sambar deer at multi-scales. The human-related disturbance would be the main factor affecting sambar deer population expansion in the future.

Keywords: Geographical information system, Habitat selection, Home range, Sambar deer, Wildlife management

Introduction

Large herbivores are primary consumers that play an important role in ecosystems and provide a substantial economic resource for many human communities. However, human land use has caused wildlife habitat loss, fragmentation, and degradation (Ceballos & Ehrlich, 2002). Furthermore, following the improvement in hunting techniques, overexploitation became the most important threat after habitat destruction for the survival of large herbivores (Groom, 2006).

Consequently, many ungulates in Asia are confined to protected areas and are limited to small populations (Baskin & Danell, 2003). Thus, conservation actions to ensure the long-term survival of these animals are required.

Space use is a fundamental component of the natural history of a species

(Patterson et al., 1999). It could be influenced by many factors: food habits, sex, mating system, metabolic need, body mass, predation, human disturbance, population density…etc (Vincent et al., 1995; Luccarini et al., 2006; Mergey et al., 2011; Walter et al., 2011). The space use of animals is usually quantified by using animal movement distances and home range sizes. Home range could be viewed as the cognitive map that an animal maintains and updates to meet its needs, including food gathering, mating, caring for the young (Moorcroft, 2012; Powell & Mitchell, 2012).

Besides understanding of animal natural history, analyses of space use attributes can be useful for further studies such as habitat selection (D’Eon & Serrouya et al., 2005), foraging behaviour (Safi et al., 2007), spatial grouping behaviour (Wagner et al., 2008;

Zhang et al., 2010), and population size estimation (Hewison et al., 2007).

Additionally, it can offer substantial information for conservation, for example, the boundary delimitation of conservation areas (Zeng et al., 2008).

Habitat selection is a process involving innate and learned behavioural decisions made by an animal about what habitat it would prefer or avoid to use (Hutto, 1985). 5

While habitat loss is the primary threat to animal, examining habitat selection is a way to evaluate the importance of habitat to species conservation. The preservation of sufficient habitat to ensure the survival of endangered species requiring conservation plans based on knowledge of habitat selection (Onorato et al., 2010). In contrast, the management of habitat under impacts of an overabundant species also requires the knowledge of habitat selection by that species (Côté et al., 2004).

Habitat selection may occur at a hierarchy of scales (Johnson, 1980; Senft et al.,

1987). Selection measured at one scale is often insufficient to predict selection at another scale (Mayor et al., 2009). Johnson (1980) defined 4 spatial scales of animal habitat selection. The 1st-order selection is defined as the geographical range of a species (geographical-range scale). The 2nd-order selection determined the home range of an individual or social group (home-range scale). The 3rd-order selection concerns the usage of various habitat components within the home range

(within-home-range scale). The 4th-order selection is the selection of food items from a feeding site. In addition to the spatial scales, the temporal scale (e.g. differences between seasonal and daily decisions) may be more important than spatial scale in some circumstances (Mayor et al., 2009). However, few studies have focused habitat selection at the temporal scale.

Study on geographical-range scale habitat selection can help us identify a species distribution, which is one of the most important factors to successfully manage and conserve a species. Nowadays, advances in habitat suitability modelling techniques combined with Geographical Information System (GIS) provide accessible and reliable tools to identify suitable habitats and to predict potential species distribution

(Anderson et al., 2003; Gavashelishvili & Javakhishvili, 2010). These tools have been applied in many studies on ungulates (e.g. Debeljak et al., 2001; Boitani et al., 2008;

Kuemmerle et al., 2010). 6

Radio telemetry is a technique popularly used in studies of animal space use and habitat selection (usually at 2nd-order and 3rd-order selection). Incorporating global positioning system (GPS) into radiocollars has revolutionized wildlife research.

Comparing with traditional radio telemetry, GPS telemetry allowed the location data to be collected in a higher spatial resolution and larger sample size (Schwartz et al.,

2009). In addition, the application of GPS telemetry can save lots of efforts in the field, especially in rugged terrain such as mountainous areas of Taiwan.

The sambar deer (Rusa unicolor) is a large ungulate species distributed throughout South and Southeast Asia (Leslie, 2011). Although this species became a pest after introduction to countries such as Australia (Gormley et al., 2011), its populations in native ranges have become small with many local-level extinctions due to extensive hunting and habitat loss (Timmins et al., 2008). Only populations in well-secured (i.e., protected or remote) areas remain abundant. Therefore, the sambar deer is listed as “vulnerable” in the IUCN Red List (Timmins et al., 2008). Although many countries have banned hunting and increased awareness about conservation issues in recent years, the recovery rate of sambar deer populations remains slow, requiring further evaluation of population distributions and dynamics (Steinmetz et al.,

2009).

The Formosan sambar deer (R. u. swinhoii) is a subspecies endemic to Taiwan

(Wilson & Reeder, 2005) and is classified as a “rare and valuable species” in the List of Protected Species in Taiwan. Before the 1990s, the number and geographic distribution of the Formosan sambar deer in Taiwan rapidly declined, reflecting similar trends recorded in other areas of sambar deer native ranges. However, in the mid-1990s, Formosan sambar deer populations in Taiwan slightly expanded (Timmins et al., 2008). Surveys such as that conducted by Wang et al. (2002) indicated that

Formosan sambar deer is mainly distributed in the Central Mountain Range of Taiwan. 7

However, as the surveys were limited by low research effort relative to the large size of the Taiwan island, this distribution pattern is a rough estimate. Consequently, due to the lack of information in many areas, the status of this sub-species of Taiwan remains unclear.

Some studies have described the habitat selection and space use of sambar deer in India (Porwal et al., 1996; Kushwaha et al., 2004; Sankar & Acharya, 2004),

Thailand (Yamada et al., 2003), Australia (Forsyth et al., 2009; Podchong et al., 2009;

Gormley et al., 2011), and USA (Richardson, 1972; Flynn et al., 1990; Shea et al.,

1990). Kushwaha et al. (2004) reported that the sambar deer had preference to high tree and herb density but low shrub cover and low direct human disturbance. Gormley et al. (2011) showed that forest cover, annual precipitation, and number of gullies were the main factors affecting sambar deer habitat suitability. Yamada et al., 2003 suggested that the geographical features were the important variables for deer distribution prediction. And the deer in Florida preferred freshwater habitats and avoids saltwater habitats (Flynn et al., 1990). In addition, Shea et al. (1990) has examined the home range and movement of this species. The mean annual home range size was 406 ha for males and 201 ha for females. Where sambar deer has been introduced in coastal Texas, annual home ranges (ha) from direct observations of known individuals were 69–124 ha for males and 38-51 ha for females (Richardson,

1972). In India, mean annual home ranges were 1,500 ha for male sambar deer and

300 ha for females (Sankar & Acharya, 2004).

However, few studies were carried out within sambar deer’s native range and for the subspecies of Taiwan. Sambar deer habitats in Taiwan are limited to areas above

1,000 m in elevation (Wang et al., 2002), the climate and vegetation types are subject to variation along the elevation gradient and are different to those of other countries.

Moreover, despite sambar deer is vulnerable in Taiwan, it has been regionally 8

overabundant at some protected areas and caused negative impacts to forest in recent years (Lee et al., 2007; Weng et al., 2009, 2010; Yen et al., 2012). Assessing what factors affect the distribution and expansion of sambar deer population by studying its space use and habitat selection will be helpful for the conservation and management of this species and its habitats.

In this thesis, I report studies using habitat suitability modelling and GPS telemetry for the Formosan sambar deer in Taiwan. The aims are to: (1) evaluate the geographical-range scale habitat selection (1st-order selection) and map the distribution pattern of Formosan sambar deer throughout Taiwan; (2) examine

Formosan sambar deer habitat selection of a home range (2nd-order selection) and within the home range (3rd-order selection), we would like to evaluate what environmental variables can affect the expansion of sambar deer; (3) estimate the habitat selection at temporal scale, we hypothesized that the sambar deer would have shorter distances from forested habitat during daytime because of their requirement of forests as shelter and cover; (4) estimate the home-range size and movement of sambar deer. The study area of the first section was the whole Taiwan. The latter three sections were conducted at Mountain Panshi, a site relatively free of human disturbance, which can represent the native behavior of sambar deer.

Methods

I. Habitat selection at geographical-range scale

1. Study area

Taiwan is located at 23° 30' N, 121° 00' E, off the southeastern coast of mainland

China (Figure 1a). The island of Taiwan is 35,801 km2. The topography of Taiwan is high and steep due to the presence of the Central Mountain Range and 4 other mountain ranges. Island elevations range from 0 to 3,952 m. The main vegetation-types change along the elevation gradients, from broadleaf forests to coniferous forests to scrub (Su, 1992). Two-thirds of the island is covered by forested mountains. Most of the coastal plains are occupied by human settlements. There are

89 protected areas, which in total cover approximately 6,951 km2 of Taiwan (Figure

1b). The climate of Taiwan is marine-tropical, with warm (mean annual temperature of the lowlands is about 23 °C) and humid (mean annual precipitation is above 2,500 mm) weather (1971–2000, Central Weather Bureau, Taiwan). However, the weather is cold in the high mountains, where it snows during the winter.

2. Data collection

The sambar deer locational data were primarily assimilated from our field observations (2008–2012). In addition, data from other field studies were also included to ensure coverage of other possible habitats (Pei et al., 2002; Pei et al., 2003;

Pei, 2004; Wu & Shi, 2006; Lee et al., 2007; Wu & Yao, 2007; C. Y. Lin, personal communication). Although the multiple datasets were collected using different techniques, thus preventing a standardized evaluation, it allowed the incorporation of data from many sites and environments in Taiwan. Two different survey techniques were used in these studies. The first technique involved the use of line transects to obtain deer presence data. Absence data was not collected because it is difficult to confirm the “absence” of a species in such surveys. The transect lines were located on 10

hiking and hunting trails. In each study site, we surveyed several transect lines to cover the different environments of that site. There were 1,582 coordinates of sambar deer tracks and signs (i.e., sightings, vocalizations, scats, footprints, tree rubbing, and shed antlers) recorded using a handheld GPS (Garmin 60CSx). In the second technique, 258 camera traps were used to obtain deer presence-absence data. Camera traps were laid 10–100 m away from the routes and were located on animal trails. The camera trap sites at which sambar deer were recorded were classified as sambar deer presence data. The camera trap sites that operated for more than 20 days without any record of sambar deer were classified as sambar-deer absence data. In total, 1,840 records of sambar deer were gathered across Taiwan, comprising 1,645 presence locations and 195 absence locations. The data covered most elevations of Taiwan

(Table 1).

These records were transformed to a resolution of 1 x 1 km grids. All records within the same grid were integrated as one presence or absence grid. A total of 361 grids of 1 km2 were sampled, representing 1% of the total land area of Taiwan. Line transect records that overlapped with absence grids were examined for cross-validation. Thereafter, 5 absence grids were revised as presence grids. Overall,

241 grids corresponded to presence records, while 120 grids corresponded to absence records.

3. Environmental variables

We used 10 environmental variables that are possibly important for sambar-deer habitat suitability (Kushwaha et al., 2004; Forsyth et al., 2009; Gormley et al., 2011).

These variables included mean elevation, standard deviation of elevation (indicator of terrain ruggedness), distance to water body, annual mean temperature, annual precipitation, vegetation type, forest area, distance to road (distance to nearest roads for vehicular traffic), road density (total length of roads in a 1 km2 grid), and human 11

settlement cover (Table 2). All the environmental variables were obtained from the ecological and environmental geographic information system database of Taiwan (Lee et al., 1997) and transformed to a resolution of 1 × 1 km by using ArcGIS 9.3 (ESRI,

Redlands, CA) and IDRISI Andes (Clark Labs, Worcester, MA, USA).

4. Statistical analyses and development of models

We divided the data into training and testing data. We used the heuristic provided by Huberty (1994) to determine the ratio of testing data to whole dataset. The heuristic is as follows:

1/[1 + √(p - 1)] where p is the number of environmental variables. Since we used 10 environmental variables, the ratio of training to testing data should be 3:1.

In this study, we selected logistic regression, discriminant analysis,

Ecological-Niche Factor Analysis (ENFA), Genetic Algorithm for Rule-set

Production (GARP), and Maximum Entropy (Maxent) to analyse sambar deer distributions. Logistic regression and discriminant analysis are presence-absence models, while the other 3 are presence-only models. These models are all regularly used for predictions of species distributions (Teixeira et al., 2001; Hirzel et al., 2002;

Brotons et al., 2004).

Logistic regression has been shown to be a powerful tool for analysing the effects of 1 or several independent variables, which are discrete or continuous, over a dichotomic (presence/absence) or polychotomic dependent variable (Hosmer &

Lemeshow, 1989). Logistic regression takes the following form:

π (x) = eg (x)/(1 + eg (x)) or π (x) = 1/(1 + e-g (x)) where π (x) represents the probability of occurrence of the target species. The g (x) is obtained by a regression equation with the form:

g (x) = β0 + β1x1 + β2x2 + … + βpxp 12

where β0 is a constant and β1, β2, … βp are the coefficients of respective independent variables x1, x2, … xp (Hosmer & Lemeshow, 1989).

Discriminant analysis is a technique used for classifying a set of observations into predefined classes that are based on a set of variables (McLachlan, 2004). Based on the observations, the technique constructs a set of linear functions of the environmental variables, which are known as discriminant functions, whereby

L = b1x1 + b2x2 + … + bnxn + c where b1, b2, … bn are discriminant coefficients; x1, x2, … xn are the environmental variables; and c is a constant. These discriminant functions are used to predict the class of a new observation with an unknown class. For a k class problem, k discriminant functions are constructed. Given a new observation, all of the k discriminant functions are evaluated, and the observation is assigned to class i if the ith discriminant function has the highest value. Logistic regression and discriminant analysis were performed with SAS 9.0 (SAS Institute Inc., Cary, NC, USA).

The principle of ENFA is to compare the distributions of the environmental variables between the presence dataset and the whole study area (Hirzel et al., 2002).

Like Principal Components Analysis, ENFA summarises several environmental variables into a few uncorrelated factors that explain most of the information. The output of ENFA includes eigenvalues and factor scores. The first factor is the marginality factor, which describes the difference between the mean habitat in the study area and species mean. The remainders are the specialisation factors, which describe how specialised the species is with reference to the available habitat range in the study area (Hirzel et al., 2002). ENFA was developed using Biomapper 4.0 (Hirzel et al., 2007). After computing the factor scores, we used the algorithm of the medians to draw a habitat-suitability map for sambar deer.

GARP is a genetic algorithm that creates ecological niche models for species 13

(Stockwell & Peters, 1999). The model describes environmental conditions under which a species should be able to maintain populations. GARP searches iteratively for non-random correlations between presence and environmental variables by using 4 types of rules: atomic, logistic regression, bioclimatic envelope, and negated bioclimatic envelope. Predicted presence is defined by these rules. We used the

Desktop GARP application (version 1.1.6; Http://www.nhm.ku.edu/desktopgarp/) and followed the normal procedure for implementation. The output of a GARP run is a binary map; hence, we applied a modification of the “best subsets” procedure described by Anderson et al. (2003). We ran 200 GARP models and selected the best

20 models that that had the highest predicted accuracy. The final GARP prediction was produced by summing the 20 selected models, which produced prediction values ranging from 0 to 20.

Maxent is a machine-learning technique that is based on the principle of maximum entropy (Pearson et al., 2004). It estimates the probability distribution of maximum entropy for each environmental variable across the study area with presence-only data (Pearson et al., 2004; Pearson et al., 2006). This distribution is calculated with the constraint that the expected value of each environmental variable under this estimated distribution matches its empirical average (Pearson et al., 2006).

Habitat suitability maps were calculated by applying Maxent models to all grids in the study area, using a logistic link function to yield probability values ranging from 0 to

1. Moreover, Maxent performs well with small sample sizes (Elith et al., 2006;

Hernandez et al., 2006). Maxent models were developed using Maxent (version 3.3.1;

Http://www.cs.princeton.edu/~schapire/maxent/).

We used different methods for each model to determine the importance of each environmental variable. The importance of each variable was determined by Wald

Chi-Square statistics in the analysis of maximum likelihood estimates. For 14

discriminant analysis, the importance of each variable was determined by standardised canonical discriminant function coefficients. For ENFA, variable importance was determined from the factor scores. For Maxent, variable importance was determined by (1) jackknife analysis of the mean gain with the training and test data, in addition to the area under the receiver operating characteristic curve (AUC); and (2) the mean percentage contribution of each environmental variable for the models (Phillips et al., 2006). Because of software restrictions, we were unable to evaluate the importance of variables for GARP.

To evaluate the performance of each model, we used the AUC (Fielding & Bell,

1997). An AUC is created by plotting the true-positive fraction against the false-positive fraction for all test points across all possible probability thresholds. The

AUC measure takes values between 0 and 1, with a value of 0.5 indicating that a model is no better than random. It is independent of prevalence and is considered as a highly effective measure for the performance of ordinal score models (Manel et al.,

2001; McPherson et al., 2004).

For conservation purposes, it is usually desirable to distinguish “suitable” from

“unsuitable” areas by setting a threshold. If the predicted probability of occurrence is larger than the threshold, then it is considered to be a prediction of presence (Pearson et al., 2004). In this study, we calculated kappa statistics under different probabilities of occurrence and selected the probability that generated the maximum kappa statistic as the threshold for each model (Freeman & Moisen, 2008).

To obtain the most robust prediction map, we used ensemble forecasting, as described by Araújo & New (2007). We calculated ensemble forecasting as weighted mean by weighting each model based on its AUC measurement (Araújo & New, 2007;

Marmion et al., 2009; Thuiller et al., 2009; Oppel et al., 2012). The habitat suitability indices of ensemble forecasting ranged from 0 to 1. We summarized the area sizes of 15

suitable habitat and optimal habitat by using 2 suitability index thresholds, that is,

0.33 and 0.67, respectively. We arbitrarily selected these 2 thresholds basing on our knowledge to Formosan sambar deer status in Taiwan.

II. Habitat selections at home-range and within-home-range scales 1. Study area

This study was conducted at the Taroko National Park, situated in Hualian,

Taichung, and Nantou counties of Taiwan. This mountainous national park encompasses an area of 920 km2, with the Central Mountain Range passing through in north-south direction and the Central-Cross-Island Highway crossing in east-west direction. The elevation ranges from sea level to 3700 m, while the area of deer tracking mostly ranges from 1,400 to 3,600 m. The mean annual temperature are 17.5

℃, 12.5 ℃, and 7.7 ℃ in elevations of 1,000 m, 2,000 m, and 3,000 m, respectively.

Areas over 3000 m in elevation snow during winter. The mean annual precipitation is over 2,000 mm. Higher participation occurs during May to October. Dominant natural vegetation at medium-to-high elevation can be roughly classified into 10 types: (1)

Yushania niitakayamensis thicket, (2) Juniperus squamata and Rhododendron pseudochrysanthum thicket, (3) Abies Kawakamii forest, (4) Tsuga formosana forest,

(5) Abies Kawakamii and Tsuga formosana forest, (6) Picea morrisonicola forest, (7)

Chamaecyparis formosana forest, (8) Pinus taiwanensis forest, (9) broadleaf-conifer mixed forest, and (10) Evergreen broadleaf forest (Yang & Hsu, 2004). The site for deer capture was Mountain Panshi, which is about 3,000 m above sea level. Although this site is only approximately 9 km away from its entrance, it actually cost two days walking to arrive due to the rugged terrain. Consequently, this site is under low human disturbance and hunting pressure.

2. Sambar deer capture and GPS data Trapping sessions were carried out between July 2009 and October 2012. We set a

1.8 m high net baited with salt. While the deer were attracted by the bait, we drove them into the net and darted them with blowpipe. The anesthetic agents used were ketamine (2.0 mg/kg) mixed with xylazine (1.2 mg/kg) (Tung et al., 1993). After anesthetized, the body measurements were taken, blood samples were collected for genetic analysis, and ectoparasites were gathered. Each animal was fitted with a GPS collar (Tellus 4D GPS collar, Followit AB, Lindesberg, Sweden; C500 GPS collar,

Tenxsys Inc., Idaho, USA; GPS GSM collar, GPS Designer company, Tainan,

Taiwan). Handling time was less than 30 min and deer were released at the location of capture. All captures were conducted with the assistance of veterinarian to minimize the stress and risk of injure to sambar deer. Most collars had a drop-off mechanism, which would be triggered by remote control or timer setting. Deer capture and handling procedures were granted by the administrations (Taroko Naitonal Park,

Forest Bureau, and animal care committee of National Taiwan Normal University).

We programmed the collars acquire locations every 1-7 h. Data were retrieved from the collars using an ultra-high frequency radio modem or from cell phone messages via Global System for Mobile Communications. We omitted any location determined by less than 3 satellites from analysis due to its low accuracy. We also visually displayed the data on Google Earth (Google, Mountain View, USA) for error examination. 3. Home-range scale habitat selection To evaluate home-range scale selection, we first defined the border of our study area as the 100% minimum convex polygon (MCP) generated by all locations of all collared deer (Vila et al., 2008). An individual deer was considered as an experiment unit. Selection was evaluated by comparison of environmental variable values

between observed and expected locations. The observed locations were random locations within each deer’s annual or seasonal home range (100% MCP), and expected locations were random locations distributed throughout the study area. We generated 1,000 random locations to represent the attributes of study area after testing a range of points (500-8000) to asses when the mean values began to stabilize. 4. Within-home-range habitat selection To test the within-home-range scale selection, we compared mean values of variables from each individual deer locations (observed locations) to mean values of an equal number of random locations within its 100% MCP home range (expected locations). The random locations are assumed to represent the habitat attributes of each home range.

5. Environmental variables and data analysis

Locations of each deer were categorized into composite and seasonal datasets.

Composite dataset included all locations of each deer. Seasonal datasets were classified by the following definition: May-October as hot/wet season and

November-April as cold/dry season. This classification was according to the monthly mean temperatures and mean precipitations data during 2009-2012 from two weather stations nearby the study site. The mean temperature and mean monthly precipitation

4.7 °C and 195 mm for cold/dry season and 10.6 °C and 273 mm for hot/wet season.

The environmental variables we examined for deer selection were slope, elevation, solar duration, aspect (i.e., compass direction), and vegetation type. We present data on proportion use of elevation, slope, and aspect. Elevation, slope, aspect, and solar duration variables are derived from ASTER Global Digital Elevation Model

(http://www.gdem.aster.ersdac.or.jp/). We reclassified aspects into 3 categories in response of different moisture gradients: mesic (338-67°), subxeric (68-157° and

248-337°), and xeric (158-247°) (Whittaker, 1960; Lai et al., 2003), hence the

expected proportions of the three categories were 25%, 50%, and 25%, respectively.

Solar duration is computed as the total hours of sun shined on each pixel (20 x 20 m) during each season. To identify the vegetation types of study area, we used a land-use map from 3rd national resource inventory conducted in 1994 by the Forest Bureau of

Taiwan. Eight habitat types are recognized as available: grassland of arrow bamboo

(Yushania niitakayamensis) (hereafter grassland), forest of Taiwan hemlock (Tsuga formosana) (hereafter hemlock forest), forest of Formosan false cypress

(Chamaecyparis taiwanensis) and Taiwan cypress (Chamaecyparis formosensis)

(hereafter cypress forest), forest of Taiwan red pine (Pinus taiwanensis) (hereafter pine forest), forest of Taiwan white fir (Abies kawakamii) (hereafter fir forest), broadleaf-conifer mixed forest (hereafter mixed forest), broadleaf forest, and open habitat (including rock lands and river bank) (Table 5).

Seasonal data sets were used for the analyses of all environmental variables except for slope because we have observed that collared deer used different areas between two seasons. Composite data set was used for the analysis of the variable “slope” because it was thought not being affected by season. The selections of slope, elevation, and solar radiation were evaluated by using t-tests between observed and expected locations (Zar 1984). Selection of aspect was assessed with Bonferroni simultaneous confidence intervals (Byers et al., 1984).

The Euclidean distance analysis was used to assess the selection of habitat type

(Gammons et al., 2009; Onorato et al., 2010). It conducted as follows. We used the

Nearest Feature Extension (Jenness Enterprises 2004) to calculate the Euclidean distance (m) of each random location and deer location to the nearest polygon of each habitat type. Points were valued as 0 m if they occurred within a habitat. To determine the habitat selection, we calculated 8 distance ratios (1 per habitat type) for each deer by dividing the average distances from observed locations by the average distances 19

from expected locations. Wilcoxon test was used to determine which habitat types were used disproportionately. Habitat types with distance ratios significantly <1 were preferred, and those significantly >1 were avoided. Habitat types were ranked by performing a series of pairwise comparisons using Mann-Whitney test.

Location data and habitat attributes were analyzed using ArcGIS 10.1 (ESRI,

Redlands, CA, USA) or ArcView 3.2 (ESRI, Redlands, CA, USA). Boundaries of study area and home ranges and random locations were generated by Geospatial

Modelling Environment (http://www.spatialecology.com/gme/index.htm). All statistics were done using SPSS 20 (SPSS Inc., Chicago, USA). We considered statistical significance for all analyses at P<0.05.

III. Habitat selection at temporal scale

The study area and data collection of this section were the same as the previous section “habitat selection at home-range and within-home-range scales“. To estimate the sambar deer habitat selection at temporal scale, we did the following analysis. The location data of collared deer were partitioned into two seasonal datasets as the previous section. We further classified the data into daytime locations and night locations. We defined the daytime for cold/dry season and hot/wet season as

6:00-17:59 and 5:00-18:59, respectively. The rest of a day was defined as night. The six forested types in the previous section (fir, hemlock, cypress, pine, mixed, broadleaf forests) were combined into “forested habitat”. We calculated the distance from the forested habitat of each location. The locations within forested habitat would have a distance of 0. We calculated the distance ratios in 2 seasons for each deer by dividing the mean distances from forested habitat in night by that in daytime. The hypothesis was that the sambar deer would have shorter distances from forested

habitat during daytime. Wilcoxon test was used to examine this prediction by testing if the distance ratios significantly >1.

IV. Space use

The study area and data collection of this section were the same as the previous section “habitat selection at home-range and within-home-range scales “. Further analyses for daily displacement and home range were estimated by using the software

R (http://www.r-project.org/) and Home Range Extension (Rodgers & Carr, 1998) in

ArcView 3.2 (ESRI, Redlands, CA, USA). To test the effects of sex and season, the data were categorized into sexes and two seasons as previous section. Annual home range was estimated only for animals tracked across 2 seasons. Mann-Whitney test was used to determine the differences between sexes and between seasons.

Home-range size was estimated using 100% MCP and 95% fixed kernel isopleth

(White & Garrott, 1990). MCP method is a traditional method which is extensively used in most animal space use studies. This method allows comparison between studies (Mergey et al., 2011). Fixed kernel has the advantage that providing information about how intensively portions of home range are used (Powell, 2000).

We used both methods to give a more complete picture of sambar deer space use

(Davini et al., 2004). Although the fix intervals ranged from 1 to 7 h, a pretest showed that the home range sizes were similar whatever using 1 or 7 h interval fixes.

The daily displacement, i.e. distance of a deer moved from one day to the next, was calculated by the following method: In each dataset (annual and seasonal datasets of each individual), we randomly chose one location of each day and computed the straight-line distances between two consecutive days. The mean daily displacement of each dataset was calculated and the outliers were excluded if the values were greater or less than three SDs from the mean. Five iterations were carried out to produce more 21

confident results. We described the mean ± SD of annual and seasonal daily displacements of all deer. In addition, we showed the maximum daily displacement as well.

Furthermore, we estimated the overlap of 100% MCP home ranges between individuals to study their interaction (Wagner et al., 2008; Zhang et al., 2010). Sa, b =

Aa, b/Aa is the proportion of animal a’s home range overlapped by animal b, where Aa represent the home range size of animal a, and Aa, b is the area size of overlap. Only pairs of individuals tracked at the same period were analyzed.

Results

I. Habitat selection at geographical-range scale

The 5 habitat suitability models (logistic regression, discriminant analysis,

ENFA, GARP, and Maxent) had AUCs of 0.894, 0.885, 0.807, 0.777, and 0.908, respectively. Each model predicted different distribution patterns for sambar deer

(Figure 2). Logistic regression, discriminant analysis, ENFA, and Maxent indicated that distance to road and the mean elevation were the most important factors for sambar deer habitat suitability (Table 3). A composite map was produced by ensemble forecasting (Figure 3a). The results showed that ensemble forecasting performs the best in all models (AUC = 0.921).

There were 7,865 grids regarded as suitable habitats for sambar in Taiwan, of which 4,464 grids were regarded as optimal habitats. Most suitable deer habitats are located in the Central Mountain Range and Xue Mountain Range. Furthermore, about

70% (5,355 of 7,865 km2) of suitable habitats are located in protected areas.

Our ensemble model indicated that sambar deer prefer habitats of medium to high elevation (above 1,500 m) and areas that are situated away from roads. The mean elevation of suitable habitats for sambar deer was about 1,500 to 2,500 m, with the predicted distribution including all areas above 3,000 m. The mean distance to roads of suitable habitats was about 5 to 12 km. In general, sambar-deer habitat suitability increased with increasing elevation and distance from roads.

There were 5 main patches of suitable habitat for sambar deer in Taiwan (Figure

3a). Two of these patches were in the Xue Mountain Range, while the other 3 were in the Central Mountain Range and Yu Mountain Range. These patches were separated by 3 major mountain highways, which were the Central Cross-Island Highway, the

Southern Cross-Island Highway, and Highway NO.7A (Figure 3a). In addition, another 3 small patches of suitable deer habitat were located in the Ali Mountain 23

Range, the Coastal Mountain Range, and the Chatianshan Protected area (Figures 1b and 3a). Overall, suitable deer habitats in areas of low elevation were scarce.

II. Habitat selections at home-range and within-home-range scales We spent 34 nights for deer capture. Twenty male and 10 female sambar deer were trapped, and 16 male and 10 female deer were collared with GPS collar. With the exception of 2 juvenile male deer, all deer captured was adults. The body measurements for adult males were: weight = 114 ± 13 kg (n = 14), shoulder height =

99 ± 8 cm (n = 18), neck circumference = 71 ± 7 cm (n = 18), and body length = 140

± 9 cm (n = 18). And that for adult females were: weight = 78 ± 12 kg (n = 9) , shoulder height = 89 ± 5 cm (n = 10), neck circumference = 47 ± 4 cm (n = 10), and body length = 130 ± 9 cm (n = 10).

Data from 12 of the collared deer have been retrieved (6 males and 6 females.

Table 4). The reasons that no data was retrieved from another 14 collars were GPS and transmitter breakdown (n = 10), collar belt breakage (n=2), and program bug (n =

2). After omitted unsuccessful locations, a total of 16,523 locations with 2,093 deer-day were used for further analyses. Two and 4 of the 12 collared deer had only tracked in the cold/dry season and hot/wet season, respectively. Thus, the sample sizes for habitat selection analyses were 10 for hot/dry season and 8 for cold/dry season. A border of study area with an area size of 3,229 ha was created by using the

16,523 locations, comprising of 8 habitat types (Figure 4, Table 5).

The deer locations had a mean slope of 27 ± 4°, while the steepest slope was 62° used by deer CL6 (Figure 5). Comparison between the two seasons showed that the deer used lower elevations during cold/dry season (2483 ± 406 m) than hot/wet season (2984 ± 222 m) (Figure 6). The use of mesic aspect (in average 40% in cold/dry season and 43% in hot/wet season) was higher than expected (25%) (Figure

7). And the use of xeric aspect during cold/dry season (12%) was much lower than expected (25%) (Figure 7). Use proportions of the eight habitat types differed between two seasons. The most frequently used types were grassland (25.4 ± 29.8%) and broadleaf forest (24.8 ± 27.0%) for cold/dry season and grassland (42.5 ± 27.3%) and hemlock forest (18.4 ± 15.7%) for hot/wet season.

In the home-range scale selection, deer’s home range environmental attributes were compared with the environmental attributes of the study area. The results suggested that the collared deer usually preferred to use areas with flatter slopes

(Table 6). In the cold/dry season, the habitat types of broadleaf forest, mixed forest, open, and hemlock forest were preferred (Table 7). And the deer tended to selected areas with higher elevation, lower solar duration, and mesic aspect, and avoided xeric aspect (Table 6). In the hot/wet season, fir forest, grassland, and hemlock forest were preferred, and cypress forest was avoided (Table 7). And the deer preferred areas with higher elevation, higher solar duration, and mesic aspect (Table 6).

In the within-home-range scale selection, environmental attributes of individual deer’s locations were compared with environmental attributes of its home range. A half of collared deer preferred to use slopes higher than expected (Table 8). In the cold/dry season, all habitat types were used in proportion to their availability (Table 9).

Higher solar duration was highly preferred. In the hot/wet season, all habitat types were used in proportion to their availability again (Table 9). And deer usually preferred higher elevations. Yet no consistent selection of solar duration was observed.

In addition, we observed no consistent selection of aspect in this scale (Table 8).

III. Habitat selection at temporal scale

In the cold/dry season, the mean distances from forested habitat of each deer ranged between 35-131 m and between 37-148 m during daytime and night, 25

respectively. Six out of the 8 deer had larger mean distances during daytime than at night (Table 10). The mean distance ratio dividing night distance by daytime distance was 1.33. There was no significant difference between the distances at daytime and at night (P = 0.069). In the hot/wet season, the mean distances from forested habitat of each deer ranged between 20-176 m and between 37-274 m during daytime and night, respectively. Nine out of the 12 deer had larger mean distances during daytime than at night (Table 10). The mean distance ratio dividing night distance by daytime distance was 1.54. The distance at daytime was significantly shorter than that at night (P =

0.012).

IV. Space use The mean annual 100% MCP home range sizes were 1078 ± 501 ha for males and 1001 ± 346 ha for females (Table 11). And the mean 95% fixed kernel home range sizes were 101 ± 24 ha for males and 102 ± 42 ha for females. No significant differences were detected between two sexes whatever estimated by 100% MCP (P =

0.299) or 95% fixed kernel (P = 0.340). And no significant differences were detected between hot/wet season and cold/dry season (100% MCP: P = 0.310; 95% fixed kernel: P = 0.310).

The mean daily displacement was 268 ± 90 m for males and 317 ± 135 m for females (Table 12). The greatest daily displacement of male was 6435 m in December

2011, and that of female was 4422 m in November 2011. Daily displacement did not differ between sexes (P = 1) and between seasons (P = 0.450).

There were 5 periods with 1, 2, 4, 3, and 2 individuals tracked, respectively

(Figure 8). The overlaps of deer home ranges were various among pairs. The proportions of overlap ranged from 0 to 80.2% (Figure 9).

Discussion

I. Habitat selection at geographical-range scale

There were about 7,865 km2 suitable habitats for sambar deer in Taiwan.

However, not all predicted habitats are populated today. For example, deer were not detected in the Ali Mountain Range (Lin, 1997) and Chatianshan Protected area

(Wang, 1994). In addition, there were large suitable habitats with small populations located in the Xue Mountain Range (Figure 3b; Yen, unpublished data). We suggested that such conditions were due to the high hunting pressure in these areas. The aboriginal tribe around these areas has a long and prevalent hunting tradition with habits which usually overexploited local wildlife resources.

The habitat suitability models indicated that distance to road and mean elevation are the 2 most important factors. This result is similar that found by Kushwaha et al.

(2004), who suggested that sambar deer avoid direct human interference, preferring areas of higher elevations. Our map derived from ensemble forecasting showed that 3 highways separate potential habitats into 5 main patches. The 3 highways were constructed about 40–50 years ago. Traffic, human settlements, lights, noise, dogs, and tourist presence along the roads cause disturbance to animals (Debeljak et al.,

2001; Klar et al., 2008). Furthermore, areas near the roads are highly vulnerable to poaching activity. Such human disturbance along roads interrupts connectivity between patches.

Furthermore, the mean elevation was another important determinant of sambar-deer habitat suitability. A previous study by Podchong et al. (2009) also indicated that geographical parameters have a great influence on sambar deer distribution predicting. This deer species prefers areas of medium to high elevation in

Taiwan. Over 90% of the optimal sambar deer habitats were located in areas of more than 1,500 m elevation. Different elevations have different climatic conditions and 27

vegetation types, which may affect habitat selection by sambar deer. However, we suggest that the effect of elevation may also be related to land exploitation at lower elevations. According to a study conducted 70 years ago by Kano (1940), sambar deer occupied ranges from 300 m to 3,600 m elevation in Taiwan. Furthermore, the bones of sambar deer have also been found at low-elevation archaeological sites in Taiwan

(Chen, 2000). Today, most areas of low elevation are exploited, and because of dense human populations, there is no intact habitat available for sambar deer in these areas.

Thus, today, sambar deer are primarily distributed in areas of medium to high elevation where human disturbance is low.

Previous research (Kushwaha et al., 2004; Gormley et al., 2011) has indicated that forest cover and annual precipitation are important determinants of sambar-deer habitat suitability. In this study, forest area was not an important factor. This might be because forest area is correlated with elevation in Taiwan. For example, as elevation increases, the proportion of forest area usually increases, while that of human settlement cover usually decreases. In addition, the arrow bamboo grasslands, which grow above the tree line, are also important habitats for sambar deer. Moreover, the study by Forsyth et al. (2009) indicated that aspect is an important predictor for sambar deer abundance. We did not use this variable in the current study because it is not suitable for use at a resolution of 1 km2 grids, which were selected due to the large

(national level) scale used here.

To obtain more reliable results of sambar deer distribution and key habitat variables, we applied 5 habitat suitability models as well as ensemble forecasting.

Each of the 5 habitat suitability models generated different predictions. These outputs show that prediction and identification of the distribution of suitable habitats for sambar deer may be influenced by model selection. Since there is uncertainty in selecting a single model that robustly predicts and identifies suitable sambar deer 28

habitats, the use of ensemble forecasting indeed improved model performance, and obtain robust predictions of species characteristics (Araújo & New, 2007; Marmion et al., 2009; Thuiller et al., 2009; Oppel et al., 2012). Our results show that ensemble forecasting performed better than all other 5 models. Although the AUC value of

Maxent was only 0.013 lower than ensemble forecasting, we thought the prediction of

Maxent was too conservative, while the prediction of ensemble forecasting was more accurate. The composite map generated from ensemble forecasting showed the most likely prediction pattern (Figure 3a).

The location data used in this study were assimilated from many independent field surveys of different areas because we wanted to analyse potential deer habitats across the maximum possible range of areas (Figure 3b). Therefore, the survey data could not be integrated into a single planned sampling design. In future, the sampling methods of Forsyth et al. (2009) and Gormley et al. (2011) should be referred to in order to improve the robustness of prediction models. However, the complicated topography and low road density of mountainous areas in Taiwan limited the application of such sampling methods in the current study. Hence, these sampling methods need to be modified according to the existing conditions in Taiwan before use in future studies.

II. Habitat selections at home-range and within-home-range scales

Our data indicated that the sambar deer were selective in their choice of habitat at both home-range scale and within-home-range scale. At home-range scale, sambar deer preferred different habitat types in two seasons and selected flatter slopes and mesic aspects in all year. At within home range scale, sambar deer preferred areas of higher solar duration during the cold/dry season.

The selections of habitat types only occurred at home range scale. Sambar locations were significantly closer than expected to broadleaf forest, mixed forest, open habitat, and hemlock forest during cold/dry season, and to fir forest, grassland, and hemlock forest during hot/wet season. The evergreen forests in medium elevations may provide sufficient food resource in winter (Lee & Lin, 2003). And the great quantities of arrow bamboo on grasslands and understory of fir forest are high nutritious foods for sambar deer when it sprouts in spring and summer (Lee &

Lin 2003; Wu & Shih, 2011). Two non-forest habitat types (open habitat in cold/dry season and grassland in hot/wet season) were also identified as favorable for sambar deer because of their nutritious grass or pasture for grazing (Semiadi et al., 1993;

Yamada et al., 2003).

The expansion of deer population could cause negative impacts to forest (Côté et al., 2004; Takatsuki, 2009). In the Yushan National Park, where probably has the most abundance sambar deer in Taiwan (Lee et al., 2007), hemlock forests suffered serious damages by sambar deer (Weng et al., 2009; 2010; 2011). Our results also showed that sambar deer preferred hemlock forest all year long. Although no selection to pine forest was found in this study, Weng et al. (2009; 2011) showed that pine forest in Yushan National Park also suffered deer damage. Stafford (1997) reported that sambar deer caused damage to pine forest in New Zealand as well. In addition, the deer locations were farther than expected to cypress forest during hot/wet season. The study of Yen et al. (2012) showed that cypress saplings were less favorable food than hemlock and spruce saplings to sambar deer in new plantations. And the study of Weng (2009; 2010) showed that the cypress forest suffered relatively low damage by sambar deer as well. Thus, we hypothesized that

the deer will utilize pine and cypress forests if the population density is over a

certain level.

Six of the 8 habitat types were selected in one or both seasons. No type was

avoided all year long. It suggested that the sambar deer is a habitat generalist. The

previous studies also reported that the sambar deer has a wide food habit and has

high variation in dietary selection depending on forage availability (Schaller 1967;

Stafford, 1997; Lee & Lin, 2003; Padmalal, 2003; Leslie, 2011). This might be the

reason why the environmental variable “vegetation type” was relatively not

important in the first-order selection. The ability of adaptation to a variety of habitats

is beneficial for the sambar deer population recovery and expansion. The future

expansion of sambar deer will likely be influenced by factors other than habitat

types。

Sambar deer occurs from sea level to alpine areas over 3,000 m above sea level throughout its native range (Leslie, 2011). The Formosan sambar deer in mountainous areas of Taiwan showed seasonal movement behavior like the sambar in the India

Himalayas (Green, 1987). The seasonal changes of food resource, temperature, and solar radiation are possible factors leading to animal seasonal movement (Zeng et al.,

2008; Jacques et al., 2009; Zeng et al., 2010). An upward migration during summer is regarded as a strategy to increase energy intake among temperate ungulates, and a downward migration during winter is considered as to increase access to food and to decrease energy consumption (Mysterud et al., 1999).

In the study of geographical-range habitat selection by sambar deer, we suggested that the deer preferred high elevations due to the lower human disturbance.

In the finer scales (2nd- and 3rd- order selections), we also found the preference to higher elevations regardless of seasons. Such selection in finer scales could be

confounded with other factors such as variation of food resource, solar radiation, and temperature.

The deer used slopes of 20-30° most frequently. Most individuals did not use slopes over 50°. And slopes over 62° were never used. These data provided the reference to studies regarding population connection and population genetics. In the

2nd-order selection, sambar deer selected flatter areas for its home range. Although a half of collared deer selected steeper slopes in the 3rd-order scale, the mean values were just slightly higher than expected values. We suggested that the selections by deer in the 3rd-order scale were mainly affected by factors other than slope. Animal habitat selections involved a trade-off among factors when their effects occur at the same scale.

The preference of higher solar duration was regarded as the main selection in the within-home-ranges scale selection in cold/ dry season, yet was not important in hot/wet season. That is because thermal balance is essential for herbivores selecting their habitats in winter (Zeng et al, 2010). Deer usually prefers areas with higher solar duration where there is more food and less snow accumulation (D’Eon, 2001; D’Eon

& Serrouya 2005). Yamada et al. (2003) also reported that the sambar deer chose areas with higher solar duration during winter in Australia. However, an adverse pattern occurred in the 2nd-order selection, i.e. selections on lower and higher solar durations during cold/dry and hot/wet seasons, respectively. We thought this adverse pattern resulted from the seasonal movement behavior. Higher elevations around mountain ridges, where were used in hot/wet season, generally receive more solar radiation than medium elevations (Zeng et al., 2010). The factors forming seasonal movement had stronger effect to sambar deer habitat selection at home-range scale and lead to this unexpected pattern.

Our study used GPS telemetry to examine factors influencing sambar deer habitat selection. However, some possible factors could not be assessed in our analysis. First, we did not examine the selection of water source, which is an essential resource to sambar deer (Flynn et al., 1990; Yamada et al., 2003; Forsyth et al., 2009,).

That is because pools at high elevations which provided water for deer are unstable and depend on rains. However, we could assess the importance of water from other variables. The highly use and preference of mesic aspect and the shorter distances from open habitat (a habitat type including river bank) in cold/dry season both indicated the deer’s need for water. Second, we observed a phenomenon that the deer was attracted by camping sites when those were used by hikers. There were about one or two groups of hikers per week passing the study area. The hikers usually arrived in the afternoon and left in the next early morning. The deer might be attracted by the human wastes and kitchen wastes of hikers due to its requirement of salt

(Matsubatashi et al., 2007). However, the frequency and number of hiker occurrence was so variable that we were not able to analyze this factor. In addition, some biotic factors such as inter-species interaction (Johnson et al., 2000; Dussault, 2005; Faas &

Weckerly, 2010), social organization, and demographic variation were not possible to analyze in current study due to a lack of related data. Furthermore, we did not consider the hunting pressure because hunting activities in our study area are few. And we did not examine the effect of predator because the only predator to sambar deer in

Taiwan, the black bear, is rare in the study area.

III. Habitat selection at temporal scale

As expected, sambar deer were located significantly closer to forested habitats during the daytime than at night. These results indicated the importance of forests as

shelter to sambar deer, even though there was little predator pressure to sambar deer in our study area. This could be a kind of innate behaviors that some ungulates maintain antipredator behavior in absence of predators (Byers 1997).

A significant pattern of diurnal difference occurred in the hot/wet season. A possible reason is that the forested habitat is important as cover to deer thermal balance in the hot season. Another possibility is related to the seasonal movement behavior and forages of sambar deer. The major foods of sambar deer were arrow bamboo at high elevations and broad leaves at medium elevations (Lee & Lin, 2003).

Thus, the deer could forage within forest at medium elevations during the cold/dry season, and left forest more often to forage on arrow bamboo during the hot/wet season. Furthermore, the aggregation of deer near the pools and the mating behavior during the rut probably contributed to this seasonal difference as well.

These results suggested that the deer habitat selections occurred at temporal scales (seasonal and diurnal) as well. The importance of habitat selection at temporal scales, which usually receives less attention than at spatial scales, should be noticed in the future.

IV. Space use

Sankar & Archarya (2004) reported that the mean annual home ranges of sambar were 1,500 ha for males and 300 ha for females in India. A more detailed study using radiotelemetry was carried out for introduced sambar deer at St. Vincent Island, USA

(Shea et al., 1990). Its results showed that the mean annual home range size was 406 ha for males and 201 ha for females. Our study provides credible data on sambar deer space use in its native range. The results suggested that the Formosan sambar deer had larger home ranges than the sambar of St. Vincent Island. We suggested that the

habitat size and elevation were the main factors contributed to such difference. The St.

Vincent Island is a subtropical island (29°N) with relatively small area size (49 km2).

In comparison, the Taroko National Park, which owns large-sized intact habitats in the

Central Mountain Range, has a more temperate-like climate because it is located at higher elevations. The small space size of St. Vincent Island could directly restrict the home range size of deer. And herbivores usually have larger home ranges in temperate zone than in tropical zone (McCullough et al., 2000). Besides, home range size is also related to body size (Endo & Doi, 1996). The larger-sized Indian sambar deer (Leslie,

2011) in St. Vincent Island is expected to have larger home ranges than the Formosan sambar deer. However, our results were contrary to this hypothesis. Thus, the habitat size and elevation might have stronger effects.

We did not found significant differences in sambar deer home range sizes between sexes and between seasons. In mammal species, body mass difference between sexes could result in the difference of home range sizes (Harestad & Bunnell,

1979). The male, who has larger body size than female, usually has larger home range to access more food. In addition, mammals may reduce their home range during winter (Jacques et al., 2009; Luccarini et al., 2006). That is because the low temperature and snow lead to higher energy expenditure in foraging. And the shortage of food decreases the efficiency of energy intake. Our results did not support these two hypotheses. It might be because the snow period of the study area is short (about

1-2 months), and the deer moved to the warmer medium-elevations where infrequently snow during winter. Furthermore, the arrow-bamboo grassland in high-elevations and the evergreen forest in medium-elevations might be able to provide sufficient food resource for both sexes of sambar deer in all seasons. Thus, little difference of home range sizes was found between sexes and between seasons.

Another reason could be that our sample size was relatively small. However, it was difficult to overcome due to the budget and the difficulties of field work.

Although there were no significant differences of home range sizes between sexes and between seasons, the mean value of the male home range size in hot/wet season was higher than that in cold/dry season. Except access to food, access to mate and rearing young are another two key factors to determine home range sizes. Male deer may rove widely to search for females or establish a small and exclusive territory during breeding season, depending on its mating strategy (Singer et al., 1981). The period between July and October is the main mating season of Formosan sambar deer

(Yang, 1988). Our result fitted for the first prediction, and was against to the second one because the sambar deer does not establish exclusive territory during mating season (Leslie, 2011; Yen & Wang, 2011). Furthermore, female deer usually reduce its home range after its fawn birth (Singer et al., 1981). The main birrthing season of captive Formosan sambar deer is between May and June (Yang, 1988), but fawns could be observed in the field from May (Yen, unpublished data) to November (Lee &

Lin, 2003). Thus, female sambar deer seasonal home ranges did not have significant difference.

The mean daily displacement of sambar deer was about 300 m. In some cases, the daily displacements were less than 10 m. No significant differences were found between sexes and between seasons. These results indicated that a deer used a stationary range for a certain period of time. In addition, the maximum daily displacements were larger than 4 km. It occurred when the deer moved in long distance between high-elevation and medium-elevation habitats at season changing periods. Since the terrain of mountainous areas in Taiwan is steep, the deer could

complete its seasonal movement in 1 or 2 days by such maximum daily displacements.

Our data showed that the overlap in home ranges of sambar deer could be as high as 80.2%. We suggested that the deer did not establish exclusive home range.

According to our field observations, the mature males were usually solitary and sometimes grouped with females during the mating season. And the females were solitary or in small groups of 2-4 deer. No exclusive territory was observed even in the mating season. Thus, a mating system of polygyny or promiscuous by sambar deer is expected. However, a study on social organization usually requires a large sample size. Our data could only provide a preliminary reference for further studies in the future.

Management implications

Not all suitable habitats are occupied by sambar deer today. While many studies recommend reintroducing target animals to the suitable but unoccupied habitats (e.g.,

Klar et al., 2008; Kuemmerle et al., 2010), it would be not appropriate in this case.

This is because sambar deer are not under immediate threat of extinction in Taiwan; hence, hasty introductions may cause unanticipated damage to the local environment

(Côté et al., 2004). Thus, we believe that monitoring the expansion process of sambar deer populations and associated environmental impacts is a more appropriate management technique at present.

With respect to the 7,865 km2 area of suitable habitats, 30% was located outside the protected areas. The largest patch, which was 260 km2, was located at Mt. Baigu, while other patches were located to the north-east of Taroko National Park, to the east of Yuli Wildlife Refuge, to the east and west of Guanshan Wildlife Refuge, and to the west of Yushan National Park (Figures 1b and 3a). We recommend that the government should consider establishing a protected area at Mt. Baigu, in parallel to expanding the ranges of other wildlife refuges and national parks. In winter of

2011-2012, 2 of the collared deer moved southward for over 8 km and exited the boundary of Taroko National Park. We lost their signals soon after the movements, yet we did not know the reason of signal lost. A guess was that they were hunted. This event may support the recommendation that natural environments adjacent to protected areas should be protected. The sambar deer is a flagship species for conservation in Taiwan; hence, the conservation of its habitats would likely benefit other large mammals, such as Reeves muntjac (Muntiacus reevesi), Formosan serow

(Capricornis swinhoei), and the black bear (Ursus thibetanus).

The habitats of sambar deer in Taiwan were divided by the mountain highways.

We hypothesized that gene flow between these patches has been limited in the recent decades. Gene flow is an important factor for conservation because the division of a species into small populations results in genetic characteristics being strongly influenced by inbreeding and genetic drift (Frankham, 1996). To address this issue, it is important to study the genetic structure of sambar deer sub-populations, and then attempt to establish connections among patches to create a viable metapopulation in the future (Kuemmerle et al., 2010; Monterrubio-Rico et al., 2010). Therefore, we recommend that a number of suitable sites that are in close vicinity to the 3 highways should be selected to monitor the process of population expansion of sambar deer.

Our study suggested that the sambar deer highly preferred to use hemlock forest.

It is consistent with the previous studies which showed a high deer damage proportion of hemlock forest (Weng et al., 2009; 2010; 2011). Thus, a long term monitoring on hemlock forests would be essential in the future. In addition, pine and cypress forests were not favorable to sambar deer in this study but were exploited in other studies

(Stafford, 1997; Weng et al., 2009; 2011). We suggested that the exploitation of pine and cypress forests by deer probably can be an indicator of deer density, i.e. the deer will use these forests more frequently when its population density is too high.

Furthermore, the previous studies on sambar deer impacts to forest in Taiwan usually focused on high elevation areas. However, we suggested that the nearby medium elevations, especially those free of human presence and lack of relative information, should be monitored in the future because we found the seasonal movement behavior of sambar deer in this study.

Conclusion

In this study, we produced a map showing the potential distribution of sambar deer in Taiwan and highlighted areas that require increased monitoring and/or protection priorities. Furthermore, we comprehensively studied habitat selection by sambar deer at multi-scales in the field by using habitat suitability models and GPS telemetry. We found that the selection of environmental factors occurred at different scales, supporting a scale-dependent hypothesis. At the geographical-range scale, human disturbance shaped the distribution pattern of the Formosan sambar deer in

Taiwan. At the home-range scale, the deer preferred flatter slops and mesic aspects in all year and selected different vegetation types and elevations at different seasons. At the within-home-range scale, solar duration might be one of deer’s main concerns.

These results suggested that the main factor affecting sambar deer expansion was human-related disturbance, not the natural environmental variables. In addition, a difference of deer habitat use between daytime and night was observed at the temporal scale, which indicated the importance of forest to the deer. Moreover, the data of sambar deer space use in its native range was established in this study. These results provide an important foundation for future studies and management implications on sambar deer in Taiwan, in addition to providing a baseline reference for other countries where this species is naturally distributed.

References

Anderson, R.P., Lew, D. & Peterson, A.T. (2003) Evaluating predictive models of

species’ distributions: criteria for selecting optimal models. Ecological

Modelling, 162, 211-232.

Araújo, M.B. & New, M. (2007) Ensemble forecasting of species distributions.

Trends in Ecology & Evolution, 22, 42-47.

Baskin, L. & Danell, K. (2003) Ecology of Ungulates - A Handbook of Species in

Eastern Europe and Northern and Central Asia, Springer, Verlag Berlin

Heidelberg, Germany.

Boitani, L., Sinibaldi, I., Corsi, F., Biase, A.D., Carranza, I.d.I., Ravagli, M., Reggiani,

G., Rondinini, C. & Trapanese, P. (2008) Distribution of medium- to

large-sized African mammals based on habitat suitability models. Biodiversity

and Conservation, 17, 605-621.

Brotons, L., Thuiller, W., Araújo, M.B. & Hirzel, A.H. (2004) Presence-absence

versus presence-only modelling methods for predicting bird habitat suitability.

Ecography, 27, 437-448.

Bryers, C. R., Steinhorst, R. K., & Krausman, P. R. (1984) Clarification if a technique

for analysis of utilization-availability data. Journal of Wildlife Management,

48, 1050-1053.

Byers, J.A. (1997) American pronghorn: social adaptations and the ghosts of

predators past, University of Chicago Press, Chicago, Illinois, USA.

Ceballos, G. & Ehrlich, P.R. (2002) Mammal population losses and the extinction

crisis. Science, 296, 904-907.

Chen, K.T. (2000) On Taiwan mammalian faunas in different periods of time and

related problems: the background materials for Taiwan zooarchaeological

studies: I. Bulletin of the Institute of History and Philology, 71, 129-198. 41

Côté, S.D., Rooney, T.P., Tremblay, J.P., Dussault, C. & Waller, D.M. (2004)

Ecological Impacts of Deer Overabundance. Annual Review of Ecology,

Evolution, and Systematics, 35, 113-147.

D'Eon, R.G. (2001) Using snow-track surveys to determine deer winter distribution

and habitat. Wildlife Society Bulletin, 29, 879-887.

D'Eon, R.G. & Serrouya, R. (2005) Mule deer seasonal movements and multiscale

resource selection using global positioning system radiotelemetry. Journal of

Mammalogy, 86, 736-744.

Davini, S., Ciuti, S., Luccarini, S. & Apollonio, M. (2004) Home range patterns of

male fallow deerDama dama in a sub-Mediterranean habitat. Acta

Theriologica, 49, 393-404.

Debeljak, M., Dzeroski, S., Jerina, K., Kobler, A. & Adamic, M. (2001) Habitat

suitability modelling for red deer (Cervus elaphus L.) in south-central

Slovenia with classification trees. Ecological Modelling, 138, 321-330.

Dussault, C., Ouellet, J.-P., Courtois, R., Huot, J., Breton, L. & Jolicoeur, H. (2005)

Linking moose habitat selection to limiting factors. Ecography, 28, 619-628.

Elith, J., Graham, C.H., Anderson, R.P., Dudík, M., Ferrier, S., Guisan, A., Hijmans,

R.J., Huettmann, F., Leathwick, J.R., Lehmann, A., Li, J., Lohmann, L.G.,

Loiselle, B.A., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton,

J.M., Peterson, A.T., Phillips, S.J., Richardson, K., Scachetti-Pereira, R.,

Schapire, R.E., Soberón, J., Williams, S., Wisz, M.S. & Zimmermann, N.E.

(2006) Novel methods improve prediction of species’ distributions from

occurrence data. Ecography, 29, 129-151.

Endo, A. & Doi, T. (1996) Home range of female sika deer Cervus nippon of Nozaki

island, the Goto Archipelago, Japan. Mammal Study, 21, 27-35.

Faas, C.J. & Weckerly, F.W. (2010) Habitat interference by axis deer on white-tailed 42

deer. Journal of Wildlife Management, 74, 698-706.

Fielding, A.H. & Bell, J.F. (1997) A review of methods for the assessment of

prediction errors in conservation presence/absence models. Environmental

Conservation, 24, 38-49.

Flynn, L.B., Shea, S.M., Lewis, J.C. & Marchinton, R.L. (1990) Ecology of sambar

deer on St. Vincent National Wildlife Refuge, Florida. Part III: Population

statistics, health and habitat use. Bulletin of Tall Timbers Research Station, 25,

63-107.

Forsyth, D.M., McLeod S.R., Scroggie M.P. & White M. (2009) Modelling the

abundance of wildlife using field surveys and GIS: non-native sambar deer

(Cervus unicolor) in the Yarra Ranges, south-eastern Australia. Wildlife

Research, 36, 231-241.

Frankham, R. (1996) Relationship of genetic variation to population size in wildlife.

Conservation Biology, 10, 1500-1508.

Freeman, E.A. & Moisen, G.G. (2008) A comparison of the performance of threshold

criteria for binary classification in terms of predicted prevalence and kappa.

Ecological Modelling, 217, 48-58.

Gammons, D.J., Mengak, M.T. & Conner, L.M. (2009) Armadillo Habitat Selection in

Southwestern Georgia. Journal of Mammalogy, 90, 356-362.

Gavashelishvili, A. & Javakhishvili, Z. (2010) Combining radio-telemetry and

random observations to model the habitat of near threatened Caucasian grouse

Tetrao mlokosiewiczi. Oryx, 44, 491-500.

Gormley, A.M., Forsyth, D.M., Griffioen, P., Lindeman, M., Ramsey, D.S.L.,

Scroggie, M.P. & Woodford, L. (2011) Using presence-only and

presence–absence data to estimate the current and potential distributions of

established invasive species. Journal of Applied Ecology, 48, 25-34. 43

Green, M.J.B. (1987) Ecological separation in Himalayan ungulates. Journal of

Zoology (London), Series B, 1, 693-719.

Groom, M.J. (2006) Threats to Biodiversity, Sunauer Associates, Sunderland, USA.

Harestad, A.S. & Bunnell, F.L. (1979) Home range and body weight - a reevaulation.

Ecology, 60, 389-402.

Hernandez, P.A., Graham, C.H., Master, L.L. & Albert, D.L. (2006) The effect of

sample size and species characteristics on performance of different species

distribution modeling methods. Ecography, 29, 773-785.

Hewison, A.J.M., Angibault, J.-M., Cargnelutti, B., Coulon, A., Rames, J.-L., Serrano,

E., Verheyden, H. & Morellet, N. (2007) Using Radio-tracking and Direct

Observation to Estimate Roe Deer Capreolus Capreolus Density in a

Fragmented Landscape: A Pilot Study. Wildelife Biology, 13, 313-320.

Hirzel, A.H., Hausser, J., Chessel, D. & Perrin, N. (2002) Ecological-niche factor

analysis: how to compute habitat-suitability maps without absence data?

Ecology, 83, 2027-2036.

Hirzel, A.H., Hausser, J. & Perrin, N. (2007) Biomapper 4.0. Department of Ecology

and Evolution, University of Lausanne, Lausanne, Switzerland.

Http://www.unil.ch/biomapper [accessed 31 August 2011].

Hosmer, D.W. & Lemeshow, S. (1989) Applied Logistic Regression, Wiley, New York,

USA.

Huberty, C.J. (1994) Applied Discriminant Analysis. Wiley, New York, USA.

Hutto, R. L. (1985) Habitat selection by nonbreeding migratory birds. P. 455-476 In:

M. L. Cody (ed.). Habitat Selection in Birds. Academic Press, Orlando, USA.

Jacques, C.N., Jenks, J.A. & Klaver, R.W. (2009) Seasonal Movements and

Home-Range Use by Female Pronghorns in Sagebrush-Steppe Communities

of Western South Dakota. Journal of Mammalogy, 90, 433-441. 44

Jenness Enterprises. 2004. Nearest features extension to Arcview. Version

3.6d. Jenness Enterprises. Flagstaff, Arizona, USA.

Johnson, B.K., Kern, J.W., Wisdom, M.J., Findholt, S.L. & Kie, J.G. (2000) Resource

selection and spatial separation of mule deer and elk during spring. Journal of

Wildlife Management, 64, 685-697.

Johnson, D. H. (1980) The comparison of usage and availability measurements for

evaluating resource preference. Ecology, 61, 65-71.

Kano, T. (1940) Zoogeographic Studies of the Tsugitaka Mountains of Formosa,

Institute for Ethnogeographical Researches, Tokyo, Japan.

Klar, N., Fernandez, N., Kramerschadt, S., Herrmann, M., Trinzen, M., Buttner, I. &

Niemitz, C. (2008) Habitat selection models for European wildcat

conservation. Biological Conservation, 141, 308-319.

Kuemmerle, T., Perzanowski, K., Chaskovskyy, O., Ostapowicz, K., Halada, L.,

Bashta, A.T., Kruhlov, I., Hostert, P., Waller, D.M. & Radeloff, V.C. (2010)

European bison habitat in the Carpathian Mountains. Biological Conservation,

143, 908-916.

Kushwaha, S.P.S., Khan, A., Habib, B., Quadri, A. & Singh, A. (2004) Evaluation of

sambar and muntjak habitats using geostatistical modelling. Current Science,

86, 1390-1400.

Lai, Y.C., Pei, J.C., Hsu, L.T. & Chiang, P.J. (2003) A preliminary study on integarting

geographic information system and multivariate statistical method to model

muntjacs habitat use in mountain area of southern Taiwan. Journal of

Photogrammetry and Remote Sensing. 8, 1-8.

Lee, L.L. & Lin, C.Y. (2003) Food habits of Formosan sambar deer (Cervus unicolor

swinhoei), Forest Bureau, Council of Agriculture, Taipei, Taiwan.

Lee, L.L., Lin, C.Y. & Chi, W.J. (2007) Monitoring of Large Mammals and 45

Population of Formosan Sambar Deer along the 2nd Section of Southern

Central Ridge Trail in Yushan National Park. Headquarter of Yushan National

Park, Nantou, Taiwan.

Lee, P.F., Liao, C.Y., Lee, Y.C., Pan, Y.H., Fu, W.H. & Chen, H.W. (1997) An

Ecological and Environmental GIS Database for Taiwan. Council of

Agriculture, Taipei, Taiwan.

Leslie, D. M. (2011) Rusa unicolor (Artiodactyla: Cervidae). Mammalian Species, 43,

1-30.

Lin, L.K. (1997) A Survey on the Fauna of the Alisan and Lulinsan Hardwood

Protected areas, Council of Agriculture, Taipei, Taiwan.

Luccarini, S., Mauri, L., Ciuti, S., Lamberti, P. & Apollonio, M. (2006) Red deer

(Cervus elaphus) spatial use in the Italian Alps: home range patterns, seasonal

migrations, and effects of snow and winter feeding. Ethology Ecology &

Evolution, 18, 127-145.

Manel, S., Williams, H.C. & Ormerod, S.J. (2001) Evaluating presence–absence

models in ecology: the need to account for prevalence. Journal of Applied

Ecology, 38, 921-931.

Marmion, M., Parviainen, M., Luoto, M., Heikkinen, R.K. & Thuiller, W. (2009)

Evaluation of consensus methods in predictive species distribution modelling.

Diversity and Distribution, 15, 59-69.

Matsubayashi, H., Lagan, P., Sukor, J.R.A. & KitayamaI, K. (2007) Seasonal and

daily use of natural licks by sambar deer (Cervus unicolor) in a Bornean

tropical rain forest. Tropics, 17, 81-86.

Mayor, S.J., Schneider, D.C., Schaefer, J.A. & Mahoney, S.P. (2009) Habitat Selection

at Multiple Scales. Ecoscience, 16, 238-247.

McCullough, D.R., Pei, K.C.J. & Wang, Y. (2000) Home Range, Activity Patterns, 46

and Habitat Relations of Reeves' Muntjacs in Taiwan Journal of Wildlife

Management, 64, 430-441.

McLachlan, G.J. (2004) Discriminant Analysis and Statistical Pattern Recognition,

Wiley, New York, USA.

McPherson, J.M., Jetz, W. & Rogers, D.J. (2004) The effects of species’ range sizes

on the accuracy of distribution models: ecological phenomenon or statistical

artefact? Journal of Applied Ecology, 41, 811-823.

Mergey, M., Helder, R. & Roeder, J.-J. (2011) Effect of forest fragmentation on

space-use patterns in the European pine marten (Martes martes). Journal of

Mammalogy, 92, 328-335.

Monterrubio-Rico, T.C., Renton, K., Ortega-Rodríguez, J.M., Pérez-Arteaga, A. &

Cancino-Murillo, R. (2010) The Endangered yellow-headed parrot Amazona

oratrix along the Pacific coast of Mexico. Oryx, 44, 602-609.

Moorcroft, P.R. (2012) Mechanistic approaches to understanding and predicting

mammalian space use: recent advances, future directions. Journal of

Mammalogy, 93, 903-916.

Mysterud, A. (1999) Seasonal migration pattern and home range of roe deer

(Capreolus capreolus) in an altitudinal gradient in southern Norway. Journal

of Zoology, 247, 479-486.

Onorato, D. P., Criffield, M., Lotz, M., Cunningham, M., McBride, R., Leone, E. H.,

Bass Jr, O. L., & Hellgren, E. C. (2010) Habitat selection by critically

endangered Florida panthers across the diel period: implications for land

management and conservation. Animal Conservation, 14, 196-205.

Oppel, S., Meirinho, A., Ramírez, I., Gardner, B., O’Connell, A.F., Miller, P.I. &

Louzao, M. (2012) Comparison of five modelling techniques to predict the

spatial distribution and abundance of seabirds. Biological Conservation, 156, 47

94-104.

Padmalal, U.K.G.K., Takatsuki, S. & Jayasekara, P. (2003) Food habits of sambar

Cervus unicolor at the Horton Plains National Park, Sri Lanka. Ecological

Research, 18, 775-782.

Patterson, B., R., Bondrup-Nielsen, S. & Messier, F. (1999) Activity patterns and

daily movements of the Eastern Coyote, Canis latrans, in Nova Scotia.

Canadian Field Naturalist, 113, 251-257.

Pearson, R.G., Dawson, T.P. & Liu, C. (2004) Modelling species distributions in

Britain: a hierarchical integration of climate and land-cover data. Ecography,

27, 285-298.

Pearson, R.G., Thuiller, W., Araújo, M.B., Martinez-Meyer, E., Brontons, L.,

McClean, C., Miles, L., Segurado, P., Dawson, T.P. & Lees, D.C. (2006)

Model-based uncertainty in species range prediction. Journal of Biogeography,

33, 1704-1711.

Pei, K.K.C. (2004) Mammal Survey in Ta-Hsueh-Shan Area, Shei-Pa National Park.

Headquarter of Shei-Pa National Park, Miaoli, Taiwan.

Pei, K.K.C., Chen, C.C. & Chen, M.T. (2003) Monitoring of Midium-to-Large Sized

Mammal Populations in Taroko National Park. Headquarter of Taroko

National Park, Hualien, Taiwan.

Pei, K.K.C., Chou, C.H., Chen, M.T., Guo, Y.L. & Liu, Y.F. (2002) Study on

Terrestrial Mammals in the Kenting National Park (III). Headquarter of

Kenting National Park, Pingtung, Taiwan.

Phillips, S.J., Anderson, R.P. & Schapire, R.E. (2006) Maximum entropy modeling of

species geographic distributions. Ecological Modelling, 190, 231-259.

Podchong, S., Schmidt-Vogt, D. & Honda, K. (2009) An improved approach for

identifying suitable habitat of Sambar Deer (Cervus unicolor Kerr) using 48

ecological niche analysis and environmental categorization: Case study at

Phu-Khieo Wildlife Sanctuary, Thailand. Ecological Modelling, 220,

2103-2114.

Porwal, M.C., Roy, P.S. & Chellamuthu, V. (1996) Wildlife habitat analysis for

'sambar' (Cervus unicolor) in Kanha National Park using remote sensing.

International Journal of Remote Sensing, 17, 2683-2697.

Powell, R. A. (2000) Animal home ranges and territories and home range estimation.

P. 65-110. In: Boitani, L. & Fuller, T. K. (eds.) Research techniques in animal

ecology: controversies and consequences. Columbia University Press, New

York, USA.

Powell, R.A. & Mitchell, M.S. (2012) What is a home range? Journal of Mammalogy,

93, 948-958.

Richardson, A.II. 1972. A natural history survey of the sambar deer (Cervus unicolor)

on the Powderhorn Ranch, Calhoun County, Texas. M.S. thesis, Texas A&M

University, College Station.

Rodgers, A.R., & Carr, A. P. (1998) HRE: The Home Range Extension for ArcView.

Ontario Ministry of Natural Resources, Centre for Northern Forest Ecosystem

Research, Thunder Bay, Ontario, Canada.

Safi, K., Konig, B. & Kerth, G. (2007) Sex differences in population genetics, home

range size and habitat use of the parti-colored bat (Vespertilio murinus,

Linnaeus 1758) in Switzerland and their consequences for conservation.

Biological Conservation, 137, 28-36.

Sankar, K. & Acharya, B. (2004) Sambar (Cervus unicolor Kerr, 1792). ENVIS

Bulletin (Wildlife Institute of India, Dehra Dun), 7, 163-170.

Schaller, G.B. (1967) The deer and the tiger: a study of wildlife in India, University of

Chicago Press, Chicago, Illinois. 49

Schwartz, C. C., Podruzny, S., Cain, S. L., & Cherry, S. (2009) Performance of spread

spectrum global positioning system collars on grizzly and black bears. The

Journal of Wildlife Management, 73, 1174-1183.

Semiadi, G., Barry, T.N. & Muir, P.D. (1993) Growth, milk intake, and behaviour of

artificially reared sambar deer (Cervus unicolor) and red deer (Cervus

elaphus). Journal of Agriculture Science, Cambridge, 121, 273-281.

Senft, R. L., Coughenour, M. B., Bailey, D. W., Rittenhouse, L. R., Sala, O. E. &

Swift, D. M. (1987) Large herbivore foraging and ecological hierarchies.

BioScience, 37: 789-799.

Shea, S.M., Flynn, L.B., Marchinton, R.L. & Lewis, J.C. (1990) Ecology of sambar

deer on St. Vincent National Wildlife Refuge, Florida. Part II: Social behavior,

movement ecology, and food habitats. Bulletin of Tall Timbers Research

Station, 25, 13-62.

Singer, F.J., Otto, D.K., Tipton, A.R. & Hable, C.P. (1981) Home ranges, movements

and habitat use of European wild boar in Tennessee. The Journal of Wildlife

Management, 45, 343-353.

Stafford, K.J. (1997) The diet and trace element status of sambar deer (Cervus

unicolor) in Manawatu district, New Zealand. New Zealand Journal of

Zoology, 24, 267-271.

Steinmetz, R., Chutipong, W., Seuaturien, N., Chirngsaard, E. & Khaengkhetkarn, M.

(2009) Population recovery patterns of Southeast Asian ungulates after

poaching. Biological Conservation, 143, 42-51.

Stockwell, D.R.B. & Peters, D. (1999) The GARP modeling system: problems and

solutions to automated spatial prediction. International Journal of

Geographical Information Science, 13, 143-158.

Su, H.J. (1992) Vegetation of Taiwan: altitudinal vegetation zones and geographical 50

climatic regions. In The Biological Resources of Taiwan: A Status Report (ed.

C.I. Peng), pp. 39-53. Institute of Botany, Academia Sinica, Taipei, Taiwan.

Takatsuki, S. (2009) Effects of sika deer on vegetation in Japan: A review. Biological

Conservation, 142, 1922-1929.

Teixeira, J., Ferrand, N. & Arntzen, J.W. (2001) Biogeography of the golden-striped

salamander Chioglossa lusitanica: a field survey and spatial modelling

approach. Ecography, 24, 618-624.

Thuiller, W., Araújo, M.B., Pearson, R.G., Whittaker, R.J., Brotons, L. & Lavorel, S.

(2004) Biodiversity conservation: uncertainty in predictions of extinction risk.

Nature, 430, 1-2.

Thuiller, W., Lafourcade, B., Engler, R. & Araújo, M.B. (2009) BIOMOD–a platform

for ensemble forecasting of species distributions. Ecography, 32, 369-373.

Timmins, R.J., Steinmetz, R., Sagar Baral, H., Samba Kumar, N., Duckworth, J.W.,

Anwarul Islam, M., Giman, B., Hedges, S., Lynam, A.J., Fellowes, J., Chan,

B.P.L. & Evans, T. (2008) Rusa unicolor. In The IUCN Red List of Threatened

Species v. 2011.1. Http://www.iucnredlist.org [accessed 31 August 2011].

Tung, K. C., Wang, K. S., Shyu, C. L., Yang, C. L., & Shih, C. H. (1993) Chemical

immobilization drug for capturing deer and its clinical application. The

Taiwan Journal of Veterinary Medicine and Animal Husbandry. 61, 9-18.

Vila, A. R., Beade, M. S., & Barrios Lamunière, D. (2008) Home range and habitat

selection of pampas deer. Journal of Zoology, 276, 95-102.

Vincent, J.P., Bideau, E., Hewison, A.J.M. & Angibault, J.M. (1995) The influence of

increasing density on body weight, kid production, home range and winter

grouping in roe deer (Capreolus capreolus). Journal of Zoology, 236, 371-382.

Wagner, A.P., Frank, L.G. & Creel, S. (2008) Spatial grouping in behaviourally

solitary striped hyaenas, Hyaena hyaena. Animal Behaviour, 75, 1131-1142. 51

Walter, W.D., Beringer, J., Hansen, L.P., Fischer, J.W., Millspaugh, J.J. & Vercauteren,

K.C. (2011) Factors affecting space use overlap by white-tailed deer in an

urban landscape. Int. J. Geogr. Inf. Sci., 25, 379-392.

Wang, Y. (1994) The fauna survey at Chatianshan Protected area. Forest Bureau,

Council of Agriculture, Taipei, Taiwan.

Wang, Y., Wang, C.C., Kuo, C.Y., Wu, H.J., Chen, S.C. & Tsai, C.C. (2002) Status of

Large Mammals in Taiwan (IV). Forest Bureau, Council of Agriculture, Taipei,

Taiwan.

Weng, G.J., Lin, C.Y., Chang, S.T. & Fan, C.H. (2010) Monitoring of Large

Mammals and Study of Population Ecology of Sambar deer in the Area of

Sinkang Mountain in the Yushan National Park, Ministry of Interior,

Headquarter of Yushan National Park, Nantou, Taiwan.

Weng, G.J., Lin, C.Y. & Tsai, C.W. (2009) Monitoring of large mammals in Singkang

Mountain and the second section of Southern Central Ridge Trail in Yushan

National Park Headquaters of Yushan national Park, Nantou, Taiwan.

Weng, G.J., Lin, C.Y. & Yang, S.Y. (2011) Surveying of the effect of sambar on the

recruitment and structure of forests in the areas of Jyunda and Guangao in the

Yushan National Park Headquaters of Yushan National Park, Nantou,

Taiwan.

White, G. C. & Garrott, R. A. (1990) Analysis of wildlife radio-tracking data.

Academic Press, New York, USA.

Whittaker, R.H. (1960) Vegetation of Siskiyou mountains, Oregon and California.

Ecological Monographs, 30.

Wilson, D. E. & Reeder, D. M. (eds.) (2005) Mammal Species of the World: A

Taxonomic and Geographic Reference. The Johns Hopkins University Press,

Baltimore, USA. 52

Wu, H.Y. & Shi, J.D. (2006) Monitoring the Artiodactyla Mammals at the Eastern

Area in Yushan National Park. Headquarter of Yushan National Park, Nantou,

Taiwan.

Wu, H.Y. & Shi, J.D. (2011) A long term ecological research on the Hehuan alpine

ecosystem (III). Headquarter of Taroko National Park, Hualian, Taiwan.

Wu, H.Y. & Yao, C.L. (2007) Monitoring Mammal Activities and Mammal-Human

Interactions in Nan-an/Bao-ai at the Eastern Area of Yushan National Park.

Headquarter of Yushan National Park, Nantou, Taiwan.

Yamada, K., Elith, J., McCarthy, M. & Zerger, A. (2003) Eliciting and integrating

expert knowledge for wildlife habitat modelling. Ecological Modelling, 165,

251-264.

Yang, S. K. (1988) The study on sambar deer restoration at Meshan, Yushan National

Park. Headquarter of Yushan National Park, Nantou, Taiwan.

Yang, Y. P. & Hsu, K. S. (2004) A preliminary study on flora diversity in alpine area

of Taroko National Park. Headquarter of Taroko National Park, Hualian,

Taiwan.

Yen, S.C., Liao, Y.C. & Wang, Y. (2012) Herbivore damage on new plantations and

surrounding forests at Danda. Quarterly Jounal of Chinese Forestry, 45(4),

1-14.

Yen, S.C. & Wang, Y. (2011) A preliminary study on the home range and social

organization of Formosan sambar deer at Mt. Chilai and Mt. Panshi

Headquarter of Taroko National Park, Hualien, Taiwan.

Yoakum, J.D. (2004) Habitat characteristics and requirements, Wildlife Management

Institute, Washington, D.C.

Zar, J.H. (1984) Biostatistical anlysis, Prentice-Hall, Englewood Cliffs, New Jersy,

USA. 53

Zeng, Z.G., Beck, P.S.A., Wang, T.J., Skidmore, A.K., Song, Y.L., Gong, H.S. & Prins,

H.H.T. (2010) Effects of Plant Phenology and Solar Radiation on Seasonal

Movement of Golden Takin in the Qinling Mountains, China. Journal of

Mammalogy, 91, 92-100.

Zeng, Z.G., Skidmore, A.K., Song, Y.L., Wang, T.J. & Gong, H.S. (2008) Seasonal

Altitudinal Movements of Golden Takin in the Qinling Mountains of China.

The Journal of Wildlife Management, 72, 611-617.

Zhang, L., Wang, Y., Zhou, Y., Newman, C., Kaneko, Y., Macdonald, D.W., Jiang, P.

& Ding, P. (2010) Ranging and Activity Patterns of the Group-Living Ferret

Badger Melogale moschata in Central China. Journal of Mammalogy, 91,

101-108.

Table 1 The elevational distribution of Formosan sambar deer (Rusa unicolor swinhoii) presence-absence records in Taiwan.

Elevation (m) Presence grid Absence gird Total

0-500 0 31 31

500-1000 9 18 27

1000-1500 42 19 61

1500-2000 37 20 57

2000-2500 37 23 60

2500-3000 57 8 65

3000-3500 59 1 60

>3500 0 0 0

Total 241 120 361

Table 2 List of environmental variables used to predict the distribution of Formosan sambar deer (Rusa unicolor swinhoii) in Taiwan. Name Values Source Mean elevation 0–3,706 m Ministry of the Interior, Taiwan Standard deviation 0–350.19 Ministry of the Interior, of elevation Taiwan Distance to water 0–8 km Institute of Transportation, body Taiwan (2008) Annual mean 6.5–25.17°C Central Weather Bureau, temperature Taiwan (1990) Annual precipitation 1,179–5,700 Central Weather Bureau, mm Taiwan (1990) Vegetation type 7 classes 1:50,000 Editorial committee of the flora of Taiwan (2nd edition 1994) Forest area 0–1 km2 1:50,000 Editorial committee of the flora of Taiwan (2nd edition 1994) Road density 0–66.60 Institute of Transportation, km/km2 Taiwan (2008) Distance to road 0–22.09 km Institute of Transportation, Taiwan (2008) Human settlement 0–1 km2 Forestry Bureau, Taiwan cover (1995)

Table 3 Ranks of gain contributions of environmental variables in the 4 habitat

suitability models: logistic regression, discriminant analysis, Ecological-Niche Factor

Analysis (ENFA), and Maximum Entropy (Maxent). The other habitat suitability

model, Genetic Algorithm for Rule-set Production, could not be used to compare the

gain contributions of each variable. ENFA could not be used to compute nominal

variables; therefore, the variable “vegetation type” was excluded from ENFA

computation.

Rank of gain Logistic Discriminant ENFA Maxentb contributions regressiona analysis

1 Distance to road Distance to road Mean elevation Mean elevation

2 Mean elevation Mean elevation Distance to road Distance to road Annual Annual mean Annual mean 3 Vegetation type precipitation temperature temperature Human Standard deviation Annual mean 4 Vegetation type settlement cover of elevation temperature Distance to water Human settlement Human settlement Annual 5 body cover cover precipitation 6 Forest area Road density Road density Forest area Annual mean Distance to water 7 Forest area Road density temperature body Annual Human settlement 8 Road density Forest area precipitation cover Standard deviation Standard deviation Distance to water Standard deviation 9 of elevation of elevation body of elevation Annual Distance to water 10 Vegetation type precipitation body a Statistically significant variables (P<0.05) were highlighted as bold. b Variables with mean percentage contribution higher than 5% were highlighted as bold.

Table 4 Data describing locations collected from Formosan sambar deer with GPS collars in Taroko National Park, Taiwan, from December 2009 to July 2013. Fix Percent Capture Tracking Successful ID Sex Weight GPS model interval fix date days fixes (h) success

CL4 F 83 Tellus 4D 2009/12/06 183 4 778 70.86

CL6 M 125 Tellus 4D 2010/7/19 356 1 8267 96.76

CL15 M 117 Tellus 4D 2010/9/16 348 4 1489 71.31

CL17 M 113 Tellus Large 2011/9/27 25 4 136 90.79

CL18 F 72 Tellus Large 2011/9/27 212 4 313 24.84

CL19 M 95 Tellus Large 2011/9/28 167 4 612 61.08

CL20 M 83 Tellus Large 2011/9/29 130 4 375 48.20

CL22 F 65 Tellus Large 2012/7/18 50 1 1151 95.92

CL23 M N/A Tellus 4D 2012/7/19 102 1 2068 84.48

CL24 F 97 Tellus Large 2012/7/21 38 1 486 53.82

CL29 F 65 GPS Designer 2012/10/24 240 7 421 51.40

CL30 F N/A GPS Designer 2012/10/25 242 7 423 50.72

Table 5 Properties of the habitat types used to analyze habitat selection by Formosan sambar deer in Taroko National Park, Taiwan, from December 2009 to July 2013. % used in % used in Mean patch Total area % of study cold/dry hot/wet size (ha) (ha) area Habitat type Description season season Overstory dominated by firs. Understory dominated by Fir forest 21 128 3.96 2.3 ± 3.9 12.6 ± 13.7 arrow bamboo and other herbs. Overstory dominated by Hemlock hemlocks. Understory 20 571 17.68 12.9 ± 11.4 18.4 ± 15.7 forest dominated by arrow bamboo and other herbs. Overstory dominated by Cypress cypress. Understory covered 36 749 23.20 13.6 ± 25.7 2.1 ± 3.6 forest by herbs and ferns. Overstory dominated by Pine forest pines. Understory having 12 831 25.74 8.7 ± 8.4 18.0 ± 15.5 some herbs and shrubs. Forest composed of coniferous and broadleaf Mixed forest 24 306 9.48 9.1 ± 16.3 0.4 ± 0.7 trees. Understory having a high variety of plants. Overstory dominated by Broadleaf broadleaf trees. Understory 21 314 9.72 24.8 ± 27.0 3.3 ± 6.5 forest having a high variety of plants. Vegetation dominated by Grassland arrow bamboo, grass, or 15 220 6.81 25.4 ± 29.8 42.5 ± 27.3 herbs. Including Rock land and Open river bank. Plants are 4 110 3.41 3.2 ± 5.2 2.7 ± 5.9 usually grass and herbs.

Table 6 Second-order habitat selection (home-range scale) by individual deer (n=12) by using landscape attributes in a Formosan sambar deer GPS telemetry study in Taroko National Park, Taiwan, from December 2009 to July 2013. “Number of deer ” reflects the number of individual deer with mean values greater than (+), less than (-), or equal to (0) the mean values of study area, based on significant t-test (for slope, elevation, and solar duration) (P < 0.05) and Bonferroni confidence interval (for aspect) between home range and study area. Preference (number of deer)

Variable + - 0 Slope 2 7 3 Elevation

Cold/dry season a 5 2 1 Hot/wet season b 10 0 0 Solar duration Cold/dry season 2 6 0 Hot/wet season 9 0 1 Aspect Cold/dry season Mesic 6 2 0 Sub 2 2 4 Xeric 1 7 0 Hot/wet season Mesic 7 1 2 Sub 3 6 1 Xeric 3 6 1 a From November to April b From May to October

Table 7 Second-order habitat selection (home-range scale) of habitat type determined by Euclidean distance analysis using location data by 12 Formosan sambar deer in Taroko National Park, Taiwan, from December 2009 to July 2013. Habitat type Ratio P value Rank Cold/dry seasona

Broadleaf forest 0.429 0.030 A Mixed forest 0.512 0.042 A Open 0.533 0.014 AB Hemlock forest 0.704 0.042 AB Fir forest 0.765 0.183 AB Grassland 0.808 0.141 B Pine forest 0.893 0.726 B Cypress forest 0.904 0.441 B Hot/wet seasonb

Fir forest 0.369 0.006 A Grassland 0.385 0.008 A Hemlock forest 0.433 0.008 A Open 0.814 0.763 B Broadleaf forest 0.902 0.415 B Mixed forest 1.088 0.610 BC Pine forest 1.145 0.359 BC Cypress forest 1.564 0.025 C Ratios <1 indicate preference, whereas ratios >1 indicate avoidance. P values <0.05 indicate the mean ratio significantly different from 1. Habitat types sharing common letter rank are similarly preferred or avoided or used in proportion to their availability. a From November to April b From May to October

Table 8 Third-order habitat selection (within-home-range sacle) by individual deer (n=12) by using landscape attributes in a Formosan sambar deer GPS telemetry study in Taroko National Park, Taiwan, from December 2009 to July 2013. “Number of deer ” reflects the number of individual deer with mean use values greater than (+), less than (-), or equal to (0) the mean value of an associated set of random locations, based on significant t-test (for slope, elevation, and solar duration) (P < 0.05) and Bonferroni confidence interval (for aspect) between use and random locations. Preference (number of deer)

Variable + - 0 Slope 7 3 2 Elevation

Cold/dry season a 5 3 0 Hot/wet season b 8 2 0 Solar duration Cold/dry season 7 1 0 Hot/wet season 4 4 2 Aspect Cold/dry season Mesic 2 5 1 Sub 3 4 1 Xeric 3 2 3 Hot/wet season Mesic 5 5 0 Sub 4 4 2 Xeric 2 4 4 a From November to April b From May to October

Table 9 Third-order habitat selection (within-home-range scale) of habitat type determined by Euclidean distance analysis using location data by 12 Formosan sambar deer in Taroko National Park, Taiwan, from December 2009 to July 2013. Habitat type Ratio P value Rank Cold/dry seasona

Pine forest 0.872 0.234 A Grassland 0.892 0.726 A Hemlock forest 0.974 0.726 A Fir forest 1.059 0.726 A Broadleaf forest 1.064 0.363 A Open 1.111 0.363 A Mixed forest 1.297 0.141 A Cypress forest 1.346 0.141 A Hot/wet seasonb

Open 0.904 0.415 A Grassland 0.917 0.641 A Fir forest 0.954 0.838 AB Broadleaf forest 1.092 0.919 AB Hemlock forest 1.151 0.359 AB Pine forest 1.149 0.415 AB Mixed forest 1.243 0.154 AB Cypress forest 1.475 0.067 B Ratios <1 indicate preference, whereas ratios >1 indicate avoidance. P values <0.05 indicate the mean ratio significantly different from 1. Habitat types sharing common letter rank are similarly preferred or avoided or used in proportion to their availability. a From November to April b From May to October

Table 10 Comparison of the distances (m) from forest pathes between daytime and night locations by 12 collared Formosan sambar deer in Taroko National Park.

Distance from forest (m) Ratio

Daytime Night (mean ± SD) (mean ± SD) Cold/dry seasona

CL4 94 ± 84 106 ± 90 1.13 CL6 37 ± 60 40 ± 63 1.08 CL15 131 ± 119 123 ± 118 0.94 CL18 87 ± 48 103 ± 49 1.18 CL19 35 ± 69 49 ± 91 1.40 CL20 71 ± 114 148 ± 156 2.08 CL29 48 ± 64 97 ± 84 2.02 CL30 45 ± 67 37 ± 61 0.82 Mean 88 69 1.13 Hot/wet seasonb

CL4 20 ± 62 40 ± 94 2.00 CL6 41 ± 67 46 ± 66 1.12 CL15 57 ± 78 81 ± 83 1.42 CL17 176 ± 165 274 ± 194 1.56 CL18 81 ± 56 78 ± 50 0.96 CL19 44 ± 80 116 ± 154 2.64 CL20 133 ± 153 265 ± 166 1.99 CL22 133 ± 162 208 ± 182 1.56 CL23 38 ± 67 51 ± 95 1.34 CL24 45 ± 65 37 ± 60 0.82 CL29 47 ± 50 96 ± 76 2.04 CL30 82 ± 85 80 ± 70 0.98 Mean 114 75 1.54 a From November to April b From May to October

Table 11 Summary statistics for 100% minimum convex polygon (MCP) and 95% fixed-kernel home ranges (ha) of Formosan sambar deer estimated by GPS telemetry in Taroko National Park, Taiwan, from December 2009 to July 2013. Cold/dry seasona Hot/wet seasonb Annual

100% MCP 95% Kernel 100% MCP 95% Kernel 100% MCP 95% Kernel (ha) (ha) (ha) (ha) (ha) (ha) Male Mean ± SD 420 ± 323 56 ± 45 744 ± 580 86 ± 48 1078 ± 501 101 ± 24 Range 117-770 15-99 98-1709 25-180 647-1780 69-122 Female Mean ± SD 621 ± 108 59 ± 57 351 ± 197 42 ± 1 1001 ± 346 102 ± 42 Range 544-697 18-99 229-578 42-43 756-1246 72-132 a From November to April b From May to October

Table 12 Summary statistics for daily displacement of Formosan sambar deer estimated by GPS telemetry in Taroko National Park, Taiwan, from December 2009 to July 2013. Daily displacement (m)

Cold/dry Hot/wet Annual seasona seasonb Male Mean ± SD 242 ± 112 314 ± 89 268 ± 90 Maximum 6435 4737 6435 Female Mean ± SD 313 ± 140 318 ± 108 317 ± 135 Maximum 4422 4391 4422 a From November to April b From May to October

Figure 1 (a) Elevation map of Taiwan, showing the locations of the mountain ranges; (b) Distribution of protected areas across Taiwan.

Figure 2 Predicted habitat of Formosan

sambar deer (Rusa unicolor swinhoii) in

Taiwan by using logistic regression,

discriminant analysis, Ecological-Niche

Factor Analysis (ENFA), Genetic

Algorithm for Rule-set Production (GARP),

and Maximum Entropy (Maxent).

Figure 3 (a) Predicted habitat of

Formosan sambar deer (Rusa

unicolor swinhoii) in Taiwan by

using ensemble forecasting. There

are 3 highways crossing the main

habitats that are suitable for sambar

deer: Central Cross-Island Highway,

Southern Cross-Island Highway, and

Highway No. 7A; (b) Recorded

locations of Formosan sambar deer

(Rusa unicolor swinhoii) in Taiwan.

Data were collected from field

surveys (2008–2011) and assimilated

previous studies (2002–2007).

Figure 4 Map of the study area showing the habitat types in the habitat selection analysis. The study area was 100% minimum convex polygon generated by all locations from 12 Formosan sambar deer tracked between December 2009 and July

2013.

Mean slope 70 62 61 62 60 51 50 50 51 49 49 47 50 47 46

)

° 40

31 31 32 32 Slope ( 30 28 27 27 26 27 24 26 20 22 17 10

0 CL4 CL6 CL15 CL17 CL18 CL19 CL20 CL22 CL23 CL24 CL29 CL30 Mean

Figure 5 The mean and range of slopes used by collared Formosan sambar deer in

Taroko National Park, Taiwan, from December 2009 to July 2013.

Cold/dry season Hot/wet season 4000 3500 3000

2500 2000

Elevation 1500 1000 500 0 CL4 CL6 CL15 CL17 CL18 CL19 CL20 CL22 CL23 CL24 CL29 CL30 Mean Formosan sambar deer

Figure 6 Mean elevations of the collared Formosan sambar deer locations during cold/dry season (November to April) and hot/wet season (May to October) in Taroko

National Park, Taiwan, from December 2009 to July 2013, shown as mean ± SD.

Cold/dry season Hot/wet season

80 70 60 50 40 30 20 10

Proportionof seasonal locations (%) 0 Mesic Subxeric Xeric

Figure 7 The proportions of aspects used by 12 Formosan sambar deer during cold/dry season (November to April) and hot/wet season (May to October) in Taroko

National Park, Taiwan, from December 2009 to July 2013, shown as mean ± SD. The aspects were reclassified into 3 moisture gradients: mesic (338-67°), subxeric

(68-157° and 248-337°), and xeric (158-247°).

Figure 8 The100% minimum convex polygon home ranges for 12 collared Formosan sambar deer in Taroko National Park. The individuals tracked at the same period were shown together at the same partition.

Overlap

100

80.2 78.7 80 72.6 70.3

60 % 44.0 41.7 40 26.0 26.7 20.8 17.3 20 15.3 15.312.5 8.1 5.0 4.1 3.6 5.7 0 0 0 0

S6, 15 S6, S15,6

S17,19 S17,20 S18,17 S18,19 S18,20 S19,17 S19,18 S19,20 S20,17 S20,18 S20,19 S22,23 S22,24 S23,22 S23,24 S24,23 S24,22 S29,30 S30,29 S17,18 Figure 9 Home range overlap proportions among collared Formosan sambar deer individuals. Overlap was computed as individual interactions (see ’Methods’).